diff --git "a/canary-1b-v2/AudioEncoder.mlmodelc/model.mil" "b/canary-1b-v2/AudioEncoder.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/canary-1b-v2/AudioEncoder.mlmodelc/model.mil" @@ -0,0 +1,6611 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "3500.14.1"}, {"coremlc-version", "3500.32.1"}})] +{ + func main(tensor input_1, tensor melspectrogram_features) { + tensor pos_emb_to_fp16 = const()[name = tensor("pos_emb_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor var_55_cast_fp16 = sub(x = pos_emb_to_fp16, y = input_1)[name = tensor("op_55_cast_fp16")]; + tensor obj_3_cast_fp16 = mul(x = var_55_cast_fp16, y = input_1)[name = tensor("obj_3_cast_fp16")]; + tensor input_1_pad_type_0 = const()[name = tensor("input_1_pad_type_0"), val = tensor("custom")]; + tensor input_1_pad_0 = const()[name = tensor("input_1_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor input_1_strides_0 = const()[name = tensor("input_1_strides_0"), val = tensor([2, 2])]; + tensor input_1_dilations_0 = const()[name = tensor("input_1_dilations_0"), val = tensor([1, 1])]; + tensor input_1_groups_0 = const()[name = tensor("input_1_groups_0"), val = tensor(1)]; + tensor pre_encode_conv_0_weight_to_fp16 = const()[name = tensor("pre_encode_conv_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768128)))]; + tensor pre_encode_conv_0_bias_to_fp16 = const()[name = tensor("pre_encode_conv_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(772800)))]; + tensor input_1_cast_fp16 = conv(bias = pre_encode_conv_0_bias_to_fp16, dilations = input_1_dilations_0, groups = input_1_groups_0, pad = input_1_pad_0, pad_type = input_1_pad_type_0, strides = input_1_strides_0, weight = pre_encode_conv_0_weight_to_fp16, x = melspectrogram_features)[name = tensor("input_1_cast_fp16")]; + tensor input_3_cast_fp16 = relu(x = input_1_cast_fp16)[name = tensor("input_3_cast_fp16")]; + tensor input_5_pad_type_0 = const()[name = tensor("input_5_pad_type_0"), val = tensor("custom")]; + tensor input_5_pad_0 = const()[name = tensor("input_5_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor input_5_strides_0 = const()[name = tensor("input_5_strides_0"), val = tensor([2, 2])]; + tensor input_5_groups_0 = const()[name = tensor("input_5_groups_0"), val = tensor(256)]; + tensor input_5_dilations_0 = const()[name = tensor("input_5_dilations_0"), val = tensor([1, 1])]; + tensor pre_encode_conv_2_weight_to_fp16 = const()[name = tensor("pre_encode_conv_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(773376)))]; + tensor pre_encode_conv_2_bias_to_fp16 = const()[name = tensor("pre_encode_conv_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(778048)))]; + tensor input_5_cast_fp16 = conv(bias = pre_encode_conv_2_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = pre_encode_conv_2_weight_to_fp16, x = input_3_cast_fp16)[name = tensor("input_5_cast_fp16")]; + tensor input_7_pad_type_0 = const()[name = tensor("input_7_pad_type_0"), val = tensor("valid")]; + tensor input_7_strides_0 = const()[name = tensor("input_7_strides_0"), val = tensor([1, 1])]; + tensor input_7_pad_0 = const()[name = tensor("input_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_7_dilations_0 = const()[name = tensor("input_7_dilations_0"), val = tensor([1, 1])]; + tensor input_7_groups_0 = const()[name = tensor("input_7_groups_0"), val = tensor(1)]; + tensor pre_encode_conv_3_weight_to_fp16 = const()[name = tensor("pre_encode_conv_3_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(778624)))]; + tensor pre_encode_conv_3_bias_to_fp16 = const()[name = tensor("pre_encode_conv_3_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(909760)))]; + tensor input_7_cast_fp16 = conv(bias = pre_encode_conv_3_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = pre_encode_conv_3_weight_to_fp16, x = input_5_cast_fp16)[name = tensor("input_7_cast_fp16")]; + tensor input_9_cast_fp16 = relu(x = input_7_cast_fp16)[name = tensor("input_9_cast_fp16")]; + tensor input_11_pad_type_0 = const()[name = tensor("input_11_pad_type_0"), val = tensor("custom")]; + tensor input_11_pad_0 = const()[name = tensor("input_11_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor input_11_strides_0 = const()[name = tensor("input_11_strides_0"), val = tensor([2, 2])]; + tensor input_11_groups_0 = const()[name = tensor("input_11_groups_0"), val = tensor(256)]; + tensor input_11_dilations_0 = const()[name = tensor("input_11_dilations_0"), val = tensor([1, 1])]; + tensor pre_encode_conv_5_weight_to_fp16 = const()[name = tensor("pre_encode_conv_5_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(910336)))]; + tensor pre_encode_conv_5_bias_to_fp16 = const()[name = tensor("pre_encode_conv_5_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(915008)))]; + tensor input_11_cast_fp16 = conv(bias = pre_encode_conv_5_bias_to_fp16, dilations = input_11_dilations_0, groups = input_11_groups_0, pad = input_11_pad_0, pad_type = input_11_pad_type_0, strides = input_11_strides_0, weight = pre_encode_conv_5_weight_to_fp16, x = input_9_cast_fp16)[name = tensor("input_11_cast_fp16")]; + tensor input_13_pad_type_0 = const()[name = tensor("input_13_pad_type_0"), val = tensor("valid")]; + tensor input_13_strides_0 = const()[name = tensor("input_13_strides_0"), val = tensor([1, 1])]; + tensor input_13_pad_0 = const()[name = tensor("input_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_13_dilations_0 = const()[name = tensor("input_13_dilations_0"), val = tensor([1, 1])]; + tensor input_13_groups_0 = const()[name = tensor("input_13_groups_0"), val = tensor(1)]; + tensor pre_encode_conv_6_weight_to_fp16 = const()[name = tensor("pre_encode_conv_6_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(915584)))]; + tensor pre_encode_conv_6_bias_to_fp16 = const()[name = tensor("pre_encode_conv_6_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1046720)))]; + tensor input_13_cast_fp16 = conv(bias = pre_encode_conv_6_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = pre_encode_conv_6_weight_to_fp16, x = input_11_cast_fp16)[name = tensor("input_13_cast_fp16")]; + tensor x_1_cast_fp16 = relu(x = input_13_cast_fp16)[name = tensor("x_1_cast_fp16")]; + tensor var_116_perm_0 = const()[name = tensor("op_116_perm_0"), val = tensor([0, 1, 3, 2])]; + tensor var_119 = const()[name = tensor("op_119"), val = tensor([1, 4096, 1, 188])]; + tensor var_116_cast_fp16 = transpose(perm = var_116_perm_0, x = x_1_cast_fp16)[name = tensor("transpose_0")]; + tensor input_15_cast_fp16 = reshape(shape = var_119, x = var_116_cast_fp16)[name = tensor("input_15_cast_fp16")]; + tensor var_129_pad_type_0 = const()[name = tensor("op_129_pad_type_0"), val = tensor("valid")]; + tensor var_129_strides_0 = const()[name = tensor("op_129_strides_0"), val = tensor([1, 1])]; + tensor var_129_pad_0 = const()[name = tensor("op_129_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_129_dilations_0 = const()[name = tensor("op_129_dilations_0"), val = tensor([1, 1])]; + tensor var_129_groups_0 = const()[name = tensor("op_129_groups_0"), val = tensor(1)]; + tensor pre_encode_out_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1047296))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4193088))), name = tensor("pre_encode_out_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor pre_encode_out_inlier_module_bias_to_fp16 = const()[name = tensor("pre_encode_out_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4193280)))]; + tensor var_129_cast_fp16 = conv(bias = pre_encode_out_inlier_module_bias_to_fp16, dilations = var_129_dilations_0, groups = var_129_groups_0, pad = var_129_pad_0, pad_type = var_129_pad_type_0, strides = var_129_strides_0, weight = pre_encode_out_inlier_module_weight_to_fp16_palettized, x = input_15_cast_fp16)[name = tensor("op_129_cast_fp16")]; + tensor var_135_pad_type_0 = const()[name = tensor("op_135_pad_type_0"), val = tensor("valid")]; + tensor var_135_strides_0 = const()[name = tensor("op_135_strides_0"), val = tensor([1, 1])]; + tensor var_135_pad_0 = const()[name = tensor("op_135_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_135_dilations_0 = const()[name = tensor("op_135_dilations_0"), val = tensor([1, 1])]; + tensor var_135_groups_0 = const()[name = tensor("op_135_groups_0"), val = tensor(1)]; + tensor pre_encode_out_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4353216))), name = tensor("pre_encode_out_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4195392))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_135_cast_fp16 = conv(dilations = var_135_dilations_0, groups = var_135_groups_0, pad = var_135_pad_0, pad_type = var_135_pad_type_0, strides = var_135_strides_0, weight = pre_encode_out_outlier_module_weight_to_fp16_sparsified, x = input_15_cast_fp16)[name = tensor("op_135_cast_fp16")]; + tensor inputs_1_cast_fp16 = add(x = var_129_cast_fp16, y = var_135_cast_fp16)[name = tensor("inputs_1_cast_fp16")]; + tensor var_141 = const()[name = tensor("op_141"), val = tensor(3)]; + tensor out_1_axes_0 = const()[name = tensor("out_1_axes_0"), val = tensor([1])]; + tensor var_172_to_fp16 = const()[name = tensor("op_172_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_172_to_fp16, x = inputs_1_cast_fp16)[name = tensor("out_1_cast_fp16")]; + tensor input_17_mean_0_to_fp16 = const()[name = tensor("input_17_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4877568)))]; + tensor input_17_variance_0_to_fp16 = const()[name = tensor("input_17_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4879680)))]; + tensor input_17_gamma_0_to_fp16 = const()[name = tensor("input_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4881792)))]; + tensor input_17_beta_0_to_fp16 = const()[name = tensor("input_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4883904)))]; + tensor input_17_epsilon_0_to_fp16 = const()[name = tensor("input_17_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_17_cast_fp16 = batch_norm(beta = input_17_beta_0_to_fp16, epsilon = input_17_epsilon_0_to_fp16, gamma = input_17_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_1_cast_fp16)[name = tensor("input_17_cast_fp16")]; + tensor var_192_pad_type_0 = const()[name = tensor("op_192_pad_type_0"), val = tensor("valid")]; + tensor var_192_strides_0 = const()[name = tensor("op_192_strides_0"), val = tensor([1, 1])]; + tensor var_192_pad_0 = const()[name = tensor("op_192_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_192_dilations_0 = const()[name = tensor("op_192_dilations_0"), val = tensor([1, 1])]; + tensor var_192_groups_0 = const()[name = tensor("op_192_groups_0"), val = tensor(1)]; + tensor layers_0_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4886016))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8031808))), name = tensor("layers_0_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8032000)))]; + tensor var_192_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_192_dilations_0, groups = var_192_groups_0, pad = var_192_pad_0, pad_type = var_192_pad_type_0, strides = var_192_strides_0, weight = layers_0_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_17_cast_fp16)[name = tensor("op_192_cast_fp16")]; + tensor var_198_pad_type_0 = const()[name = tensor("op_198_pad_type_0"), val = tensor("valid")]; + tensor var_198_strides_0 = const()[name = tensor("op_198_strides_0"), val = tensor([1, 1])]; + tensor var_198_pad_0 = const()[name = tensor("op_198_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_198_dilations_0 = const()[name = tensor("op_198_dilations_0"), val = tensor([1, 1])]; + tensor var_198_groups_0 = const()[name = tensor("op_198_groups_0"), val = tensor(1)]; + tensor layers_0_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8200384))), name = tensor("layers_0_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8040256))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_198_cast_fp16 = conv(dilations = var_198_dilations_0, groups = var_198_groups_0, pad = var_198_pad_0, pad_type = var_198_pad_type_0, strides = var_198_strides_0, weight = layers_0_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_17_cast_fp16)[name = tensor("op_198_cast_fp16")]; + tensor input_19_cast_fp16 = add(x = var_192_cast_fp16, y = var_198_cast_fp16)[name = tensor("input_19_cast_fp16")]; + tensor input_21_cast_fp16 = silu(x = input_19_cast_fp16)[name = tensor("input_21_cast_fp16")]; + tensor var_209_pad_type_0 = const()[name = tensor("op_209_pad_type_0"), val = tensor("valid")]; + tensor var_209_strides_0 = const()[name = tensor("op_209_strides_0"), val = tensor([1, 1])]; + tensor var_209_pad_0 = const()[name = tensor("op_209_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_209_dilations_0 = const()[name = tensor("op_209_dilations_0"), val = tensor([1, 1])]; + tensor var_209_groups_0 = const()[name = tensor("op_209_groups_0"), val = tensor(1)]; + tensor layers_0_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8724736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11870528))), name = tensor("layers_0_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_209_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_209_dilations_0, groups = var_209_groups_0, pad = var_209_pad_0, pad_type = var_209_pad_type_0, strides = var_209_strides_0, weight = layers_0_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_21_cast_fp16)[name = tensor("op_209_cast_fp16")]; + tensor var_215_pad_type_0 = const()[name = tensor("op_215_pad_type_0"), val = tensor("valid")]; + tensor var_215_strides_0 = const()[name = tensor("op_215_strides_0"), val = tensor([1, 1])]; + tensor var_215_pad_0 = const()[name = tensor("op_215_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_215_dilations_0 = const()[name = tensor("op_215_dilations_0"), val = tensor([1, 1])]; + tensor var_215_groups_0 = const()[name = tensor("op_215_groups_0"), val = tensor(1)]; + tensor layers_0_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12026048))), name = tensor("layers_0_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11870720))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_215_cast_fp16 = conv(dilations = var_215_dilations_0, groups = var_215_groups_0, pad = var_215_pad_0, pad_type = var_215_pad_type_0, strides = var_215_strides_0, weight = layers_0_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_21_cast_fp16)[name = tensor("op_215_cast_fp16")]; + tensor x_3_cast_fp16 = add(x = var_209_cast_fp16, y = var_215_cast_fp16)[name = tensor("x_3_cast_fp16")]; + tensor var_217_to_fp16 = const()[name = tensor("op_217_to_fp16"), val = tensor(0x1p-1)]; + tensor var_218_cast_fp16 = mul(x = x_3_cast_fp16, y = var_217_to_fp16)[name = tensor("op_218_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = var_218_cast_fp16)[name = tensor("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = tensor("out_3_axes_0"), val = tensor([1])]; + tensor var_228_to_fp16 = const()[name = tensor("op_228_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_228_to_fp16, x = inputs_3_cast_fp16)[name = tensor("out_3_cast_fp16")]; + tensor obj_1_gamma_0_to_fp16 = const()[name = tensor("obj_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12550400)))]; + tensor obj_1_beta_0_to_fp16 = const()[name = tensor("obj_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12552512)))]; + tensor obj_1_epsilon_0_to_fp16 = const()[name = tensor("obj_1_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_3_cast_fp16)[name = tensor("obj_1_cast_fp16")]; + tensor var_253_pad_type_0 = const()[name = tensor("op_253_pad_type_0"), val = tensor("valid")]; + tensor var_253_strides_0 = const()[name = tensor("op_253_strides_0"), val = tensor([1, 1])]; + tensor var_253_pad_0 = const()[name = tensor("op_253_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_253_dilations_0 = const()[name = tensor("op_253_dilations_0"), val = tensor([1, 1])]; + tensor var_253_groups_0 = const()[name = tensor("op_253_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12554624))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13341120))), name = tensor("layers_0_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_253_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_253_dilations_0, groups = var_253_groups_0, pad = var_253_pad_0, pad_type = var_253_pad_type_0, strides = var_253_strides_0, weight = layers_0_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = tensor("op_253_cast_fp16")]; + tensor var_259_pad_type_0 = const()[name = tensor("op_259_pad_type_0"), val = tensor("valid")]; + tensor var_259_strides_0 = const()[name = tensor("op_259_strides_0"), val = tensor([1, 1])]; + tensor var_259_pad_0 = const()[name = tensor("op_259_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_259_dilations_0 = const()[name = tensor("op_259_dilations_0"), val = tensor([1, 1])]; + tensor var_259_groups_0 = const()[name = tensor("op_259_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13379776))), name = tensor("layers_0_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13341312))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_259_cast_fp16 = conv(dilations = var_259_dilations_0, groups = var_259_groups_0, pad = var_259_pad_0, pad_type = var_259_pad_type_0, strides = var_259_strides_0, weight = layers_0_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_1_cast_fp16)[name = tensor("op_259_cast_fp16")]; + tensor query_1_cast_fp16 = add(x = var_253_cast_fp16, y = var_259_cast_fp16)[name = tensor("query_1_cast_fp16")]; + tensor var_268_pad_type_0 = const()[name = tensor("op_268_pad_type_0"), val = tensor("valid")]; + tensor var_268_strides_0 = const()[name = tensor("op_268_strides_0"), val = tensor([1, 1])]; + tensor var_268_pad_0 = const()[name = tensor("op_268_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_268_dilations_0 = const()[name = tensor("op_268_dilations_0"), val = tensor([1, 1])]; + tensor var_268_groups_0 = const()[name = tensor("op_268_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13510912))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14297408))), name = tensor("layers_0_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_268_cast_fp16 = conv(dilations = var_268_dilations_0, groups = var_268_groups_0, pad = var_268_pad_0, pad_type = var_268_pad_type_0, strides = var_268_strides_0, weight = layers_0_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = tensor("op_268_cast_fp16")]; + tensor var_274_pad_type_0 = const()[name = tensor("op_274_pad_type_0"), val = tensor("valid")]; + tensor var_274_strides_0 = const()[name = tensor("op_274_strides_0"), val = tensor([1, 1])]; + tensor var_274_pad_0 = const()[name = tensor("op_274_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_274_dilations_0 = const()[name = tensor("op_274_dilations_0"), val = tensor([1, 1])]; + tensor var_274_groups_0 = const()[name = tensor("op_274_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14344128))), name = tensor("layers_0_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14297600))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_274_cast_fp16 = conv(dilations = var_274_dilations_0, groups = var_274_groups_0, pad = var_274_pad_0, pad_type = var_274_pad_type_0, strides = var_274_strides_0, weight = layers_0_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_1_cast_fp16)[name = tensor("op_274_cast_fp16")]; + tensor key_1_cast_fp16 = add(x = var_268_cast_fp16, y = var_274_cast_fp16)[name = tensor("key_1_cast_fp16")]; + tensor var_284_pad_type_0 = const()[name = tensor("op_284_pad_type_0"), val = tensor("valid")]; + tensor var_284_strides_0 = const()[name = tensor("op_284_strides_0"), val = tensor([1, 1])]; + tensor var_284_pad_0 = const()[name = tensor("op_284_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_284_dilations_0 = const()[name = tensor("op_284_dilations_0"), val = tensor([1, 1])]; + tensor var_284_groups_0 = const()[name = tensor("op_284_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14475264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15261760))), name = tensor("layers_0_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_284_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_284_dilations_0, groups = var_284_groups_0, pad = var_284_pad_0, pad_type = var_284_pad_type_0, strides = var_284_strides_0, weight = layers_0_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = tensor("op_284_cast_fp16")]; + tensor var_290_pad_type_0 = const()[name = tensor("op_290_pad_type_0"), val = tensor("valid")]; + tensor var_290_strides_0 = const()[name = tensor("op_290_strides_0"), val = tensor([1, 1])]; + tensor var_290_pad_0 = const()[name = tensor("op_290_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_290_dilations_0 = const()[name = tensor("op_290_dilations_0"), val = tensor([1, 1])]; + tensor var_290_groups_0 = const()[name = tensor("op_290_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15304448))), name = tensor("layers_0_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15261952))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_290_cast_fp16 = conv(dilations = var_290_dilations_0, groups = var_290_groups_0, pad = var_290_pad_0, pad_type = var_290_pad_type_0, strides = var_290_strides_0, weight = layers_0_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_1_cast_fp16)[name = tensor("op_290_cast_fp16")]; + tensor value_1_cast_fp16 = add(x = var_284_cast_fp16, y = var_290_cast_fp16)[name = tensor("value_1_cast_fp16")]; + tensor var_293_to_fp16 = const()[name = tensor("op_293_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15435584)))]; + tensor query_3_cast_fp16 = add(x = query_1_cast_fp16, y = var_293_to_fp16)[name = tensor("query_3_cast_fp16")]; + tensor var_296_to_fp16 = const()[name = tensor("op_296_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15437696)))]; + tensor q_with_bias_v_1_cast_fp16 = add(x = query_1_cast_fp16, y = var_296_to_fp16)[name = tensor("q_with_bias_v_1_cast_fp16")]; + tensor var_306_pad_type_0 = const()[name = tensor("op_306_pad_type_0"), val = tensor("valid")]; + tensor var_306_strides_0 = const()[name = tensor("op_306_strides_0"), val = tensor([1, 1])]; + tensor var_306_pad_0 = const()[name = tensor("op_306_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_306_dilations_0 = const()[name = tensor("op_306_dilations_0"), val = tensor([1, 1])]; + tensor var_306_groups_0 = const()[name = tensor("op_306_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15439808))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16226304))), name = tensor("layers_0_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_306_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_306_dilations_0, groups = var_306_groups_0, pad = var_306_pad_0, pad_type = var_306_pad_type_0, strides = var_306_strides_0, weight = layers_0_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_306_cast_fp16")]; + tensor var_312_pad_type_0 = const()[name = tensor("op_312_pad_type_0"), val = tensor("valid")]; + tensor var_312_strides_0 = const()[name = tensor("op_312_strides_0"), val = tensor([1, 1])]; + tensor var_312_pad_0 = const()[name = tensor("op_312_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_312_dilations_0 = const()[name = tensor("op_312_dilations_0"), val = tensor([1, 1])]; + tensor var_312_groups_0 = const()[name = tensor("op_312_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16308416))), name = tensor("layers_0_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16226496))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_312_cast_fp16 = conv(dilations = var_312_dilations_0, groups = var_312_groups_0, pad = var_312_pad_0, pad_type = var_312_pad_type_0, strides = var_312_strides_0, weight = layers_0_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_312_cast_fp16")]; + tensor p_1_cast_fp16 = add(x = var_306_cast_fp16, y = var_312_cast_fp16)[name = tensor("p_1_cast_fp16")]; + tensor var_316 = const()[name = tensor("op_316"), val = tensor([1, 8, 128, 188])]; + tensor var_317_cast_fp16 = reshape(shape = var_316, x = q_with_bias_v_1_cast_fp16)[name = tensor("op_317_cast_fp16")]; + tensor var_318 = const()[name = tensor("op_318"), val = tensor([1, 8, 128, -1])]; + tensor var_319_cast_fp16 = reshape(shape = var_318, x = p_1_cast_fp16)[name = tensor("op_319_cast_fp16")]; + tensor matrix_bd_1_transpose_x_0 = const()[name = tensor("matrix_bd_1_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_1_transpose_y_0 = const()[name = tensor("matrix_bd_1_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_1_cast_fp16 = matmul(transpose_x = matrix_bd_1_transpose_x_0, transpose_y = matrix_bd_1_transpose_y_0, x = var_317_cast_fp16, y = var_319_cast_fp16)[name = tensor("matrix_bd_1_cast_fp16")]; + tensor matrix_bd_3_pad_0 = const()[name = tensor("matrix_bd_3_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_3_mode_0 = const()[name = tensor("matrix_bd_3_mode_0"), val = tensor("constant")]; + tensor const_10_to_fp16 = const()[name = tensor("const_10_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_3_cast_fp16 = pad(constant_val = const_10_to_fp16, mode = matrix_bd_3_mode_0, pad = matrix_bd_3_pad_0, x = matrix_bd_1_cast_fp16)[name = tensor("matrix_bd_3_cast_fp16")]; + tensor var_328 = const()[name = tensor("op_328"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_5_cast_fp16 = reshape(shape = var_328, x = matrix_bd_3_cast_fp16)[name = tensor("matrix_bd_5_cast_fp16")]; + tensor var_332_begin_0 = const()[name = tensor("op_332_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_332_end_0 = const()[name = tensor("op_332_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_332_end_mask_0 = const()[name = tensor("op_332_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_332_cast_fp16 = slice_by_index(begin = var_332_begin_0, end = var_332_end_0, end_mask = var_332_end_mask_0, x = matrix_bd_5_cast_fp16)[name = tensor("op_332_cast_fp16")]; + tensor var_333 = const()[name = tensor("op_333"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_7_cast_fp16 = reshape(shape = var_333, x = var_332_cast_fp16)[name = tensor("matrix_bd_7_cast_fp16")]; + tensor var_338_begin_0 = const()[name = tensor("op_338_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_338_end_0 = const()[name = tensor("op_338_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_338_end_mask_0 = const()[name = tensor("op_338_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_338_cast_fp16 = slice_by_index(begin = var_338_begin_0, end = var_338_end_0, end_mask = var_338_end_mask_0, x = matrix_bd_7_cast_fp16)[name = tensor("op_338_cast_fp16")]; + tensor var_339_to_fp16 = const()[name = tensor("op_339_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_1_cast_fp16 = mul(x = var_338_cast_fp16, y = var_339_to_fp16)[name = tensor("qk_mask_1_cast_fp16")]; + tensor var_343 = const()[name = tensor("op_343"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_1_cast_fp16 = reshape(shape = var_343, x = query_3_cast_fp16)[name = tensor("mh_q_1_cast_fp16")]; + tensor var_345_to_fp16 = const()[name = tensor("op_345_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_346_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_345_to_fp16)[name = tensor("op_346_cast_fp16")]; + tensor var_349 = const()[name = tensor("op_349"), val = tensor([1, 8, 128, 188])]; + tensor var_350_cast_fp16 = reshape(shape = var_349, x = key_1_cast_fp16)[name = tensor("op_350_cast_fp16")]; + tensor mh_w_1_transpose_x_0 = const()[name = tensor("mh_w_1_transpose_x_0"), val = tensor(true)]; + tensor mh_w_1_transpose_y_0 = const()[name = tensor("mh_w_1_transpose_y_0"), val = tensor(false)]; + tensor mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_346_cast_fp16, y = var_350_cast_fp16)[name = tensor("mh_w_1_cast_fp16")]; + tensor mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = qk_mask_1_cast_fp16)[name = tensor("mh_w_3_cast_fp16")]; + tensor var_354_cast_fp16 = softmax(axis = var_141, x = mh_w_3_cast_fp16)[name = tensor("op_354_cast_fp16")]; + tensor var_355 = const()[name = tensor("op_355"), val = tensor([1, 8, 128, 188])]; + tensor var_356_cast_fp16 = reshape(shape = var_355, x = value_1_cast_fp16)[name = tensor("op_356_cast_fp16")]; + tensor attn_1_transpose_x_0 = const()[name = tensor("attn_1_transpose_x_0"), val = tensor(false)]; + tensor attn_1_transpose_y_0 = const()[name = tensor("attn_1_transpose_y_0"), val = tensor(true)]; + tensor attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_356_cast_fp16, y = var_354_cast_fp16)[name = tensor("attn_1_cast_fp16")]; + tensor var_359 = const()[name = tensor("op_359"), val = tensor([1, 1024, 1, 188])]; + tensor input_23_cast_fp16 = reshape(shape = var_359, x = attn_1_cast_fp16)[name = tensor("input_23_cast_fp16")]; + tensor var_369_pad_type_0 = const()[name = tensor("op_369_pad_type_0"), val = tensor("valid")]; + tensor var_369_strides_0 = const()[name = tensor("op_369_strides_0"), val = tensor([1, 1])]; + tensor var_369_pad_0 = const()[name = tensor("op_369_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_369_dilations_0 = const()[name = tensor("op_369_dilations_0"), val = tensor([1, 1])]; + tensor var_369_groups_0 = const()[name = tensor("op_369_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16439552))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17226048))), name = tensor("layers_0_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_369_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_369_dilations_0, groups = var_369_groups_0, pad = var_369_pad_0, pad_type = var_369_pad_type_0, strides = var_369_strides_0, weight = layers_0_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_23_cast_fp16)[name = tensor("op_369_cast_fp16")]; + tensor var_375_pad_type_0 = const()[name = tensor("op_375_pad_type_0"), val = tensor("valid")]; + tensor var_375_strides_0 = const()[name = tensor("op_375_strides_0"), val = tensor([1, 1])]; + tensor var_375_pad_0 = const()[name = tensor("op_375_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_375_dilations_0 = const()[name = tensor("op_375_dilations_0"), val = tensor([1, 1])]; + tensor var_375_groups_0 = const()[name = tensor("op_375_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17265408))), name = tensor("layers_0_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17226240))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_375_cast_fp16 = conv(dilations = var_375_dilations_0, groups = var_375_groups_0, pad = var_375_pad_0, pad_type = var_375_pad_type_0, strides = var_375_strides_0, weight = layers_0_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_23_cast_fp16)[name = tensor("op_375_cast_fp16")]; + tensor obj_5_cast_fp16 = add(x = var_369_cast_fp16, y = var_375_cast_fp16)[name = tensor("obj_5_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_5_cast_fp16)[name = tensor("inputs_5_cast_fp16")]; + tensor out_5_axes_0 = const()[name = tensor("out_5_axes_0"), val = tensor([1])]; + tensor var_386_to_fp16 = const()[name = tensor("op_386_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_386_to_fp16, x = inputs_5_cast_fp16)[name = tensor("out_5_cast_fp16")]; + tensor input_25_gamma_0_to_fp16 = const()[name = tensor("input_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17396544)))]; + tensor input_25_beta_0_to_fp16 = const()[name = tensor("input_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17398656)))]; + tensor input_25_epsilon_0_to_fp16 = const()[name = tensor("input_25_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_25_cast_fp16 = batch_norm(beta = input_25_beta_0_to_fp16, epsilon = input_25_epsilon_0_to_fp16, gamma = input_25_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_5_cast_fp16)[name = tensor("input_25_cast_fp16")]; + tensor var_407_pad_type_0 = const()[name = tensor("op_407_pad_type_0"), val = tensor("valid")]; + tensor var_407_strides_0 = const()[name = tensor("op_407_strides_0"), val = tensor([1, 1])]; + tensor var_407_pad_0 = const()[name = tensor("op_407_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_407_dilations_0 = const()[name = tensor("op_407_dilations_0"), val = tensor([1, 1])]; + tensor var_407_groups_0 = const()[name = tensor("op_407_groups_0"), val = tensor(1)]; + tensor layers_0_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17400768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18973696))), name = tensor("layers_0_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_407_cast_fp16 = conv(dilations = var_407_dilations_0, groups = var_407_groups_0, pad = var_407_pad_0, pad_type = var_407_pad_type_0, strides = var_407_strides_0, weight = layers_0_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = tensor("op_407_cast_fp16")]; + tensor var_413_pad_type_0 = const()[name = tensor("op_413_pad_type_0"), val = tensor("valid")]; + tensor var_413_strides_0 = const()[name = tensor("op_413_strides_0"), val = tensor([1, 1])]; + tensor var_413_pad_0 = const()[name = tensor("op_413_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_413_dilations_0 = const()[name = tensor("op_413_dilations_0"), val = tensor([1, 1])]; + tensor var_413_groups_0 = const()[name = tensor("op_413_groups_0"), val = tensor(1)]; + tensor layers_0_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19043904))), name = tensor("layers_0_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18973888))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_413_cast_fp16 = conv(dilations = var_413_dilations_0, groups = var_413_groups_0, pad = var_413_pad_0, pad_type = var_413_pad_type_0, strides = var_413_strides_0, weight = layers_0_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_25_cast_fp16)[name = tensor("op_413_cast_fp16")]; + tensor input_27_cast_fp16 = add(x = var_407_cast_fp16, y = var_413_cast_fp16)[name = tensor("input_27_cast_fp16")]; + tensor input_29_split_num_splits_0 = const()[name = tensor("input_29_split_num_splits_0"), val = tensor(2)]; + tensor input_29_split_axis_0 = const()[name = tensor("input_29_split_axis_0"), val = tensor(1)]; + tensor input_29_split_cast_fp16_0, tensor input_29_split_cast_fp16_1 = split(axis = input_29_split_axis_0, num_splits = input_29_split_num_splits_0, x = input_27_cast_fp16)[name = tensor("input_29_split_cast_fp16")]; + tensor input_29_split_1_sigmoid_cast_fp16 = sigmoid(x = input_29_split_cast_fp16_1)[name = tensor("input_29_split_1_sigmoid_cast_fp16")]; + tensor input_29_cast_fp16 = mul(x = input_29_split_cast_fp16_0, y = input_29_split_1_sigmoid_cast_fp16)[name = tensor("input_29_cast_fp16")]; + tensor input_31_pad_type_0 = const()[name = tensor("input_31_pad_type_0"), val = tensor("custom")]; + tensor input_31_pad_0 = const()[name = tensor("input_31_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_31_groups_0 = const()[name = tensor("input_31_groups_0"), val = tensor(1024)]; + tensor input_31_strides_0 = const()[name = tensor("input_31_strides_0"), val = tensor([1, 1])]; + tensor input_31_dilations_0 = const()[name = tensor("input_31_dilations_0"), val = tensor([1, 1])]; + tensor const_268_to_fp16 = const()[name = tensor("const_268_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19306112)))]; + tensor const_269_to_fp16 = const()[name = tensor("const_269_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19324608)))]; + tensor input_33_cast_fp16 = conv(bias = const_269_to_fp16, dilations = input_31_dilations_0, groups = input_31_groups_0, pad = input_31_pad_0, pad_type = input_31_pad_type_0, strides = input_31_strides_0, weight = const_268_to_fp16, x = input_29_cast_fp16)[name = tensor("input_33_cast_fp16")]; + tensor input_35_cast_fp16 = silu(x = input_33_cast_fp16)[name = tensor("input_35_cast_fp16")]; + tensor var_435_pad_type_0 = const()[name = tensor("op_435_pad_type_0"), val = tensor("valid")]; + tensor var_435_strides_0 = const()[name = tensor("op_435_strides_0"), val = tensor([1, 1])]; + tensor var_435_pad_0 = const()[name = tensor("op_435_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_435_dilations_0 = const()[name = tensor("op_435_dilations_0"), val = tensor([1, 1])]; + tensor var_435_groups_0 = const()[name = tensor("op_435_groups_0"), val = tensor(1)]; + tensor layers_0_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19326720))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20113216))), name = tensor("layers_0_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_435_cast_fp16 = conv(dilations = var_435_dilations_0, groups = var_435_groups_0, pad = var_435_pad_0, pad_type = var_435_pad_type_0, strides = var_435_strides_0, weight = layers_0_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_35_cast_fp16)[name = tensor("op_435_cast_fp16")]; + tensor var_441_pad_type_0 = const()[name = tensor("op_441_pad_type_0"), val = tensor("valid")]; + tensor var_441_strides_0 = const()[name = tensor("op_441_strides_0"), val = tensor([1, 1])]; + tensor var_441_pad_0 = const()[name = tensor("op_441_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_441_dilations_0 = const()[name = tensor("op_441_dilations_0"), val = tensor([1, 1])]; + tensor var_441_groups_0 = const()[name = tensor("op_441_groups_0"), val = tensor(1)]; + tensor layers_0_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20148160))), name = tensor("layers_0_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20113408))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_441_cast_fp16 = conv(dilations = var_441_dilations_0, groups = var_441_groups_0, pad = var_441_pad_0, pad_type = var_441_pad_type_0, strides = var_441_strides_0, weight = layers_0_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_35_cast_fp16)[name = tensor("op_441_cast_fp16")]; + tensor x_5_cast_fp16 = add(x = var_435_cast_fp16, y = var_441_cast_fp16)[name = tensor("x_5_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = x_5_cast_fp16)[name = tensor("inputs_7_cast_fp16")]; + tensor out_7_axes_0 = const()[name = tensor("out_7_axes_0"), val = tensor([1])]; + tensor var_452_to_fp16 = const()[name = tensor("op_452_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_452_to_fp16, x = inputs_7_cast_fp16)[name = tensor("out_7_cast_fp16")]; + tensor input_37_gamma_0_to_fp16 = const()[name = tensor("input_37_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20279296)))]; + tensor input_37_beta_0_to_fp16 = const()[name = tensor("input_37_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20281408)))]; + tensor input_37_epsilon_0_to_fp16 = const()[name = tensor("input_37_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_37_cast_fp16 = batch_norm(beta = input_37_beta_0_to_fp16, epsilon = input_37_epsilon_0_to_fp16, gamma = input_37_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_7_cast_fp16)[name = tensor("input_37_cast_fp16")]; + tensor var_472_pad_type_0 = const()[name = tensor("op_472_pad_type_0"), val = tensor("valid")]; + tensor var_472_strides_0 = const()[name = tensor("op_472_strides_0"), val = tensor([1, 1])]; + tensor var_472_pad_0 = const()[name = tensor("op_472_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_472_dilations_0 = const()[name = tensor("op_472_dilations_0"), val = tensor([1, 1])]; + tensor var_472_groups_0 = const()[name = tensor("op_472_groups_0"), val = tensor(1)]; + tensor layers_0_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20283520))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23429312))), name = tensor("layers_0_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_472_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_472_dilations_0, groups = var_472_groups_0, pad = var_472_pad_0, pad_type = var_472_pad_type_0, strides = var_472_strides_0, weight = layers_0_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_37_cast_fp16)[name = tensor("op_472_cast_fp16")]; + tensor var_478_pad_type_0 = const()[name = tensor("op_478_pad_type_0"), val = tensor("valid")]; + tensor var_478_strides_0 = const()[name = tensor("op_478_strides_0"), val = tensor([1, 1])]; + tensor var_478_pad_0 = const()[name = tensor("op_478_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_478_dilations_0 = const()[name = tensor("op_478_dilations_0"), val = tensor([1, 1])]; + tensor var_478_groups_0 = const()[name = tensor("op_478_groups_0"), val = tensor(1)]; + tensor layers_0_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23574656))), name = tensor("layers_0_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23429504))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_478_cast_fp16 = conv(dilations = var_478_dilations_0, groups = var_478_groups_0, pad = var_478_pad_0, pad_type = var_478_pad_type_0, strides = var_478_strides_0, weight = layers_0_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_37_cast_fp16)[name = tensor("op_478_cast_fp16")]; + tensor input_39_cast_fp16 = add(x = var_472_cast_fp16, y = var_478_cast_fp16)[name = tensor("input_39_cast_fp16")]; + tensor input_41_cast_fp16 = silu(x = input_39_cast_fp16)[name = tensor("input_41_cast_fp16")]; + tensor var_489_pad_type_0 = const()[name = tensor("op_489_pad_type_0"), val = tensor("valid")]; + tensor var_489_strides_0 = const()[name = tensor("op_489_strides_0"), val = tensor([1, 1])]; + tensor var_489_pad_0 = const()[name = tensor("op_489_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_489_dilations_0 = const()[name = tensor("op_489_dilations_0"), val = tensor([1, 1])]; + tensor var_489_groups_0 = const()[name = tensor("op_489_groups_0"), val = tensor(1)]; + tensor layers_0_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24099008))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27244800))), name = tensor("layers_0_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_489_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_489_dilations_0, groups = var_489_groups_0, pad = var_489_pad_0, pad_type = var_489_pad_type_0, strides = var_489_strides_0, weight = layers_0_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_41_cast_fp16)[name = tensor("op_489_cast_fp16")]; + tensor var_495_pad_type_0 = const()[name = tensor("op_495_pad_type_0"), val = tensor("valid")]; + tensor var_495_strides_0 = const()[name = tensor("op_495_strides_0"), val = tensor([1, 1])]; + tensor var_495_pad_0 = const()[name = tensor("op_495_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_495_dilations_0 = const()[name = tensor("op_495_dilations_0"), val = tensor([1, 1])]; + tensor var_495_groups_0 = const()[name = tensor("op_495_groups_0"), val = tensor(1)]; + tensor layers_0_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27384896))), name = tensor("layers_0_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27244992))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_495_cast_fp16 = conv(dilations = var_495_dilations_0, groups = var_495_groups_0, pad = var_495_pad_0, pad_type = var_495_pad_type_0, strides = var_495_strides_0, weight = layers_0_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_41_cast_fp16)[name = tensor("op_495_cast_fp16")]; + tensor x_7_cast_fp16 = add(x = var_489_cast_fp16, y = var_495_cast_fp16)[name = tensor("x_7_cast_fp16")]; + tensor var_497_to_fp16 = const()[name = tensor("op_497_to_fp16"), val = tensor(0x1p-1)]; + tensor var_498_cast_fp16 = mul(x = x_7_cast_fp16, y = var_497_to_fp16)[name = tensor("op_498_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = var_498_cast_fp16)[name = tensor("inputs_9_cast_fp16")]; + tensor out_9_axes_0 = const()[name = tensor("out_9_axes_0"), val = tensor([1])]; + tensor var_508_to_fp16 = const()[name = tensor("op_508_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_508_to_fp16, x = inputs_9_cast_fp16)[name = tensor("out_9_cast_fp16")]; + tensor inputs_11_gamma_0_to_fp16 = const()[name = tensor("inputs_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27909248)))]; + tensor inputs_11_beta_0_to_fp16 = const()[name = tensor("inputs_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27911360)))]; + tensor inputs_11_epsilon_0_to_fp16 = const()[name = tensor("inputs_11_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_11_cast_fp16 = batch_norm(beta = inputs_11_beta_0_to_fp16, epsilon = inputs_11_epsilon_0_to_fp16, gamma = inputs_11_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_9_cast_fp16)[name = tensor("inputs_11_cast_fp16")]; + tensor var_522 = const()[name = tensor("op_522"), val = tensor(3)]; + tensor out_11_axes_0 = const()[name = tensor("out_11_axes_0"), val = tensor([1])]; + tensor var_553_to_fp16 = const()[name = tensor("op_553_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_553_to_fp16, x = inputs_11_cast_fp16)[name = tensor("out_11_cast_fp16")]; + tensor input_43_gamma_0_to_fp16 = const()[name = tensor("input_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27913472)))]; + tensor input_43_beta_0_to_fp16 = const()[name = tensor("input_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27915584)))]; + tensor input_43_epsilon_0_to_fp16 = const()[name = tensor("input_43_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_43_cast_fp16 = batch_norm(beta = input_43_beta_0_to_fp16, epsilon = input_43_epsilon_0_to_fp16, gamma = input_43_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_11_cast_fp16)[name = tensor("input_43_cast_fp16")]; + tensor var_573_pad_type_0 = const()[name = tensor("op_573_pad_type_0"), val = tensor("valid")]; + tensor var_573_strides_0 = const()[name = tensor("op_573_strides_0"), val = tensor([1, 1])]; + tensor var_573_pad_0 = const()[name = tensor("op_573_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_573_dilations_0 = const()[name = tensor("op_573_dilations_0"), val = tensor([1, 1])]; + tensor var_573_groups_0 = const()[name = tensor("op_573_groups_0"), val = tensor(1)]; + tensor layers_1_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27917696))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(31063488))), name = tensor("layers_1_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_573_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_573_dilations_0, groups = var_573_groups_0, pad = var_573_pad_0, pad_type = var_573_pad_type_0, strides = var_573_strides_0, weight = layers_1_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_43_cast_fp16)[name = tensor("op_573_cast_fp16")]; + tensor var_579_pad_type_0 = const()[name = tensor("op_579_pad_type_0"), val = tensor("valid")]; + tensor var_579_strides_0 = const()[name = tensor("op_579_strides_0"), val = tensor([1, 1])]; + tensor var_579_pad_0 = const()[name = tensor("op_579_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_579_dilations_0 = const()[name = tensor("op_579_dilations_0"), val = tensor([1, 1])]; + tensor var_579_groups_0 = const()[name = tensor("op_579_groups_0"), val = tensor(1)]; + tensor layers_1_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(31214016))), name = tensor("layers_1_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(31063680))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_579_cast_fp16 = conv(dilations = var_579_dilations_0, groups = var_579_groups_0, pad = var_579_pad_0, pad_type = var_579_pad_type_0, strides = var_579_strides_0, weight = layers_1_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_43_cast_fp16)[name = tensor("op_579_cast_fp16")]; + tensor input_45_cast_fp16 = add(x = var_573_cast_fp16, y = var_579_cast_fp16)[name = tensor("input_45_cast_fp16")]; + tensor input_47_cast_fp16 = silu(x = input_45_cast_fp16)[name = tensor("input_47_cast_fp16")]; + tensor var_590_pad_type_0 = const()[name = tensor("op_590_pad_type_0"), val = tensor("valid")]; + tensor var_590_strides_0 = const()[name = tensor("op_590_strides_0"), val = tensor([1, 1])]; + tensor var_590_pad_0 = const()[name = tensor("op_590_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_590_dilations_0 = const()[name = tensor("op_590_dilations_0"), val = tensor([1, 1])]; + tensor var_590_groups_0 = const()[name = tensor("op_590_groups_0"), val = tensor(1)]; + tensor layers_1_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(31738368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34884160))), name = tensor("layers_1_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_590_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_590_dilations_0, groups = var_590_groups_0, pad = var_590_pad_0, pad_type = var_590_pad_type_0, strides = var_590_strides_0, weight = layers_1_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_47_cast_fp16)[name = tensor("op_590_cast_fp16")]; + tensor var_596_pad_type_0 = const()[name = tensor("op_596_pad_type_0"), val = tensor("valid")]; + tensor var_596_strides_0 = const()[name = tensor("op_596_strides_0"), val = tensor([1, 1])]; + tensor var_596_pad_0 = const()[name = tensor("op_596_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_596_dilations_0 = const()[name = tensor("op_596_dilations_0"), val = tensor([1, 1])]; + tensor var_596_groups_0 = const()[name = tensor("op_596_groups_0"), val = tensor(1)]; + tensor layers_1_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35068544))), name = tensor("layers_1_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34884352))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_596_cast_fp16 = conv(dilations = var_596_dilations_0, groups = var_596_groups_0, pad = var_596_pad_0, pad_type = var_596_pad_type_0, strides = var_596_strides_0, weight = layers_1_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_47_cast_fp16)[name = tensor("op_596_cast_fp16")]; + tensor x_9_cast_fp16 = add(x = var_590_cast_fp16, y = var_596_cast_fp16)[name = tensor("x_9_cast_fp16")]; + tensor var_598_to_fp16 = const()[name = tensor("op_598_to_fp16"), val = tensor(0x1p-1)]; + tensor var_599_cast_fp16 = mul(x = x_9_cast_fp16, y = var_598_to_fp16)[name = tensor("op_599_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = var_599_cast_fp16)[name = tensor("inputs_13_cast_fp16")]; + tensor out_13_axes_0 = const()[name = tensor("out_13_axes_0"), val = tensor([1])]; + tensor var_609_to_fp16 = const()[name = tensor("op_609_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_609_to_fp16, x = inputs_13_cast_fp16)[name = tensor("out_13_cast_fp16")]; + tensor obj_7_gamma_0_to_fp16 = const()[name = tensor("obj_7_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35592896)))]; + tensor obj_7_beta_0_to_fp16 = const()[name = tensor("obj_7_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35595008)))]; + tensor obj_7_epsilon_0_to_fp16 = const()[name = tensor("obj_7_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_7_cast_fp16 = batch_norm(beta = obj_7_beta_0_to_fp16, epsilon = obj_7_epsilon_0_to_fp16, gamma = obj_7_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_13_cast_fp16)[name = tensor("obj_7_cast_fp16")]; + tensor var_634_pad_type_0 = const()[name = tensor("op_634_pad_type_0"), val = tensor("valid")]; + tensor var_634_strides_0 = const()[name = tensor("op_634_strides_0"), val = tensor([1, 1])]; + tensor var_634_pad_0 = const()[name = tensor("op_634_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_634_dilations_0 = const()[name = tensor("op_634_dilations_0"), val = tensor([1, 1])]; + tensor var_634_groups_0 = const()[name = tensor("op_634_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35597120))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36383616))), name = tensor("layers_1_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_634_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_634_dilations_0, groups = var_634_groups_0, pad = var_634_pad_0, pad_type = var_634_pad_type_0, strides = var_634_strides_0, weight = layers_1_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_7_cast_fp16)[name = tensor("op_634_cast_fp16")]; + tensor var_640_pad_type_0 = const()[name = tensor("op_640_pad_type_0"), val = tensor("valid")]; + tensor var_640_strides_0 = const()[name = tensor("op_640_strides_0"), val = tensor([1, 1])]; + tensor var_640_pad_0 = const()[name = tensor("op_640_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_640_dilations_0 = const()[name = tensor("op_640_dilations_0"), val = tensor([1, 1])]; + tensor var_640_groups_0 = const()[name = tensor("op_640_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36420992))), name = tensor("layers_1_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36383808))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_640_cast_fp16 = conv(dilations = var_640_dilations_0, groups = var_640_groups_0, pad = var_640_pad_0, pad_type = var_640_pad_type_0, strides = var_640_strides_0, weight = layers_1_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_7_cast_fp16)[name = tensor("op_640_cast_fp16")]; + tensor query_5_cast_fp16 = add(x = var_634_cast_fp16, y = var_640_cast_fp16)[name = tensor("query_5_cast_fp16")]; + tensor var_649_pad_type_0 = const()[name = tensor("op_649_pad_type_0"), val = tensor("valid")]; + tensor var_649_strides_0 = const()[name = tensor("op_649_strides_0"), val = tensor([1, 1])]; + tensor var_649_pad_0 = const()[name = tensor("op_649_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_649_dilations_0 = const()[name = tensor("op_649_dilations_0"), val = tensor([1, 1])]; + tensor var_649_groups_0 = const()[name = tensor("op_649_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36552128))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37338624))), name = tensor("layers_1_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_649_cast_fp16 = conv(dilations = var_649_dilations_0, groups = var_649_groups_0, pad = var_649_pad_0, pad_type = var_649_pad_type_0, strides = var_649_strides_0, weight = layers_1_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_7_cast_fp16)[name = tensor("op_649_cast_fp16")]; + tensor var_655_pad_type_0 = const()[name = tensor("op_655_pad_type_0"), val = tensor("valid")]; + tensor var_655_strides_0 = const()[name = tensor("op_655_strides_0"), val = tensor([1, 1])]; + tensor var_655_pad_0 = const()[name = tensor("op_655_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_655_dilations_0 = const()[name = tensor("op_655_dilations_0"), val = tensor([1, 1])]; + tensor var_655_groups_0 = const()[name = tensor("op_655_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37385792))), name = tensor("layers_1_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37338816))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_655_cast_fp16 = conv(dilations = var_655_dilations_0, groups = var_655_groups_0, pad = var_655_pad_0, pad_type = var_655_pad_type_0, strides = var_655_strides_0, weight = layers_1_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_7_cast_fp16)[name = tensor("op_655_cast_fp16")]; + tensor key_3_cast_fp16 = add(x = var_649_cast_fp16, y = var_655_cast_fp16)[name = tensor("key_3_cast_fp16")]; + tensor var_665_pad_type_0 = const()[name = tensor("op_665_pad_type_0"), val = tensor("valid")]; + tensor var_665_strides_0 = const()[name = tensor("op_665_strides_0"), val = tensor([1, 1])]; + tensor var_665_pad_0 = const()[name = tensor("op_665_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_665_dilations_0 = const()[name = tensor("op_665_dilations_0"), val = tensor([1, 1])]; + tensor var_665_groups_0 = const()[name = tensor("op_665_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37516928))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38303424))), name = tensor("layers_1_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_665_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_665_dilations_0, groups = var_665_groups_0, pad = var_665_pad_0, pad_type = var_665_pad_type_0, strides = var_665_strides_0, weight = layers_1_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_7_cast_fp16)[name = tensor("op_665_cast_fp16")]; + tensor var_671_pad_type_0 = const()[name = tensor("op_671_pad_type_0"), val = tensor("valid")]; + tensor var_671_strides_0 = const()[name = tensor("op_671_strides_0"), val = tensor([1, 1])]; + tensor var_671_pad_0 = const()[name = tensor("op_671_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_671_dilations_0 = const()[name = tensor("op_671_dilations_0"), val = tensor([1, 1])]; + tensor var_671_groups_0 = const()[name = tensor("op_671_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38338944))), name = tensor("layers_1_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38303616))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_671_cast_fp16 = conv(dilations = var_671_dilations_0, groups = var_671_groups_0, pad = var_671_pad_0, pad_type = var_671_pad_type_0, strides = var_671_strides_0, weight = layers_1_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_7_cast_fp16)[name = tensor("op_671_cast_fp16")]; + tensor value_3_cast_fp16 = add(x = var_665_cast_fp16, y = var_671_cast_fp16)[name = tensor("value_3_cast_fp16")]; + tensor var_674_to_fp16 = const()[name = tensor("op_674_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38470080)))]; + tensor query_7_cast_fp16 = add(x = query_5_cast_fp16, y = var_674_to_fp16)[name = tensor("query_7_cast_fp16")]; + tensor var_677_to_fp16 = const()[name = tensor("op_677_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38472192)))]; + tensor q_with_bias_v_3_cast_fp16 = add(x = query_5_cast_fp16, y = var_677_to_fp16)[name = tensor("q_with_bias_v_3_cast_fp16")]; + tensor var_687_pad_type_0 = const()[name = tensor("op_687_pad_type_0"), val = tensor("valid")]; + tensor var_687_strides_0 = const()[name = tensor("op_687_strides_0"), val = tensor([1, 1])]; + tensor var_687_pad_0 = const()[name = tensor("op_687_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_687_dilations_0 = const()[name = tensor("op_687_dilations_0"), val = tensor([1, 1])]; + tensor var_687_groups_0 = const()[name = tensor("op_687_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38474304))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39260800))), name = tensor("layers_1_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_687_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_687_dilations_0, groups = var_687_groups_0, pad = var_687_pad_0, pad_type = var_687_pad_type_0, strides = var_687_strides_0, weight = layers_1_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_687_cast_fp16")]; + tensor var_693_pad_type_0 = const()[name = tensor("op_693_pad_type_0"), val = tensor("valid")]; + tensor var_693_strides_0 = const()[name = tensor("op_693_strides_0"), val = tensor([1, 1])]; + tensor var_693_pad_0 = const()[name = tensor("op_693_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_693_dilations_0 = const()[name = tensor("op_693_dilations_0"), val = tensor([1, 1])]; + tensor var_693_groups_0 = const()[name = tensor("op_693_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39340032))), name = tensor("layers_1_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39260992))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_693_cast_fp16 = conv(dilations = var_693_dilations_0, groups = var_693_groups_0, pad = var_693_pad_0, pad_type = var_693_pad_type_0, strides = var_693_strides_0, weight = layers_1_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_693_cast_fp16")]; + tensor p_3_cast_fp16 = add(x = var_687_cast_fp16, y = var_693_cast_fp16)[name = tensor("p_3_cast_fp16")]; + tensor var_697 = const()[name = tensor("op_697"), val = tensor([1, 8, 128, 188])]; + tensor var_698_cast_fp16 = reshape(shape = var_697, x = q_with_bias_v_3_cast_fp16)[name = tensor("op_698_cast_fp16")]; + tensor var_699 = const()[name = tensor("op_699"), val = tensor([1, 8, 128, -1])]; + tensor var_700_cast_fp16 = reshape(shape = var_699, x = p_3_cast_fp16)[name = tensor("op_700_cast_fp16")]; + tensor matrix_bd_9_transpose_x_0 = const()[name = tensor("matrix_bd_9_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_9_transpose_y_0 = const()[name = tensor("matrix_bd_9_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_9_cast_fp16 = matmul(transpose_x = matrix_bd_9_transpose_x_0, transpose_y = matrix_bd_9_transpose_y_0, x = var_698_cast_fp16, y = var_700_cast_fp16)[name = tensor("matrix_bd_9_cast_fp16")]; + tensor matrix_bd_11_pad_0 = const()[name = tensor("matrix_bd_11_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_11_mode_0 = const()[name = tensor("matrix_bd_11_mode_0"), val = tensor("constant")]; + tensor const_21_to_fp16 = const()[name = tensor("const_21_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_11_cast_fp16 = pad(constant_val = const_21_to_fp16, mode = matrix_bd_11_mode_0, pad = matrix_bd_11_pad_0, x = matrix_bd_9_cast_fp16)[name = tensor("matrix_bd_11_cast_fp16")]; + tensor var_709 = const()[name = tensor("op_709"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_13_cast_fp16 = reshape(shape = var_709, x = matrix_bd_11_cast_fp16)[name = tensor("matrix_bd_13_cast_fp16")]; + tensor var_713_begin_0 = const()[name = tensor("op_713_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_713_end_0 = const()[name = tensor("op_713_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_713_end_mask_0 = const()[name = tensor("op_713_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_713_cast_fp16 = slice_by_index(begin = var_713_begin_0, end = var_713_end_0, end_mask = var_713_end_mask_0, x = matrix_bd_13_cast_fp16)[name = tensor("op_713_cast_fp16")]; + tensor var_714 = const()[name = tensor("op_714"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_15_cast_fp16 = reshape(shape = var_714, x = var_713_cast_fp16)[name = tensor("matrix_bd_15_cast_fp16")]; + tensor var_719_begin_0 = const()[name = tensor("op_719_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_719_end_0 = const()[name = tensor("op_719_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_719_end_mask_0 = const()[name = tensor("op_719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_719_cast_fp16 = slice_by_index(begin = var_719_begin_0, end = var_719_end_0, end_mask = var_719_end_mask_0, x = matrix_bd_15_cast_fp16)[name = tensor("op_719_cast_fp16")]; + tensor var_720_to_fp16 = const()[name = tensor("op_720_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_3_cast_fp16 = mul(x = var_719_cast_fp16, y = var_720_to_fp16)[name = tensor("qk_mask_3_cast_fp16")]; + tensor var_724 = const()[name = tensor("op_724"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_3_cast_fp16 = reshape(shape = var_724, x = query_7_cast_fp16)[name = tensor("mh_q_3_cast_fp16")]; + tensor var_726_to_fp16 = const()[name = tensor("op_726_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_727_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_726_to_fp16)[name = tensor("op_727_cast_fp16")]; + tensor var_730 = const()[name = tensor("op_730"), val = tensor([1, 8, 128, 188])]; + tensor var_731_cast_fp16 = reshape(shape = var_730, x = key_3_cast_fp16)[name = tensor("op_731_cast_fp16")]; + tensor mh_w_5_transpose_x_0 = const()[name = tensor("mh_w_5_transpose_x_0"), val = tensor(true)]; + tensor mh_w_5_transpose_y_0 = const()[name = tensor("mh_w_5_transpose_y_0"), val = tensor(false)]; + tensor mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_727_cast_fp16, y = var_731_cast_fp16)[name = tensor("mh_w_5_cast_fp16")]; + tensor mh_w_7_cast_fp16 = add(x = mh_w_5_cast_fp16, y = qk_mask_3_cast_fp16)[name = tensor("mh_w_7_cast_fp16")]; + tensor var_735_cast_fp16 = softmax(axis = var_522, x = mh_w_7_cast_fp16)[name = tensor("op_735_cast_fp16")]; + tensor var_736 = const()[name = tensor("op_736"), val = tensor([1, 8, 128, 188])]; + tensor var_737_cast_fp16 = reshape(shape = var_736, x = value_3_cast_fp16)[name = tensor("op_737_cast_fp16")]; + tensor attn_3_transpose_x_0 = const()[name = tensor("attn_3_transpose_x_0"), val = tensor(false)]; + tensor attn_3_transpose_y_0 = const()[name = tensor("attn_3_transpose_y_0"), val = tensor(true)]; + tensor attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_737_cast_fp16, y = var_735_cast_fp16)[name = tensor("attn_3_cast_fp16")]; + tensor var_740 = const()[name = tensor("op_740"), val = tensor([1, 1024, 1, 188])]; + tensor input_49_cast_fp16 = reshape(shape = var_740, x = attn_3_cast_fp16)[name = tensor("input_49_cast_fp16")]; + tensor var_750_pad_type_0 = const()[name = tensor("op_750_pad_type_0"), val = tensor("valid")]; + tensor var_750_strides_0 = const()[name = tensor("op_750_strides_0"), val = tensor([1, 1])]; + tensor var_750_pad_0 = const()[name = tensor("op_750_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_750_dilations_0 = const()[name = tensor("op_750_dilations_0"), val = tensor([1, 1])]; + tensor var_750_groups_0 = const()[name = tensor("op_750_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39471168))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40257664))), name = tensor("layers_1_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_750_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_750_dilations_0, groups = var_750_groups_0, pad = var_750_pad_0, pad_type = var_750_pad_type_0, strides = var_750_strides_0, weight = layers_1_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_49_cast_fp16)[name = tensor("op_750_cast_fp16")]; + tensor var_756_pad_type_0 = const()[name = tensor("op_756_pad_type_0"), val = tensor("valid")]; + tensor var_756_strides_0 = const()[name = tensor("op_756_strides_0"), val = tensor([1, 1])]; + tensor var_756_pad_0 = const()[name = tensor("op_756_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_756_dilations_0 = const()[name = tensor("op_756_dilations_0"), val = tensor([1, 1])]; + tensor var_756_groups_0 = const()[name = tensor("op_756_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40292224))), name = tensor("layers_1_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40257856))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_756_cast_fp16 = conv(dilations = var_756_dilations_0, groups = var_756_groups_0, pad = var_756_pad_0, pad_type = var_756_pad_type_0, strides = var_756_strides_0, weight = layers_1_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_49_cast_fp16)[name = tensor("op_756_cast_fp16")]; + tensor obj_9_cast_fp16 = add(x = var_750_cast_fp16, y = var_756_cast_fp16)[name = tensor("obj_9_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_9_cast_fp16)[name = tensor("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = tensor("out_15_axes_0"), val = tensor([1])]; + tensor var_767_to_fp16 = const()[name = tensor("op_767_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_767_to_fp16, x = inputs_15_cast_fp16)[name = tensor("out_15_cast_fp16")]; + tensor input_51_gamma_0_to_fp16 = const()[name = tensor("input_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40423360)))]; + tensor input_51_beta_0_to_fp16 = const()[name = tensor("input_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40425472)))]; + tensor input_51_epsilon_0_to_fp16 = const()[name = tensor("input_51_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_51_cast_fp16 = batch_norm(beta = input_51_beta_0_to_fp16, epsilon = input_51_epsilon_0_to_fp16, gamma = input_51_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_15_cast_fp16)[name = tensor("input_51_cast_fp16")]; + tensor var_788_pad_type_0 = const()[name = tensor("op_788_pad_type_0"), val = tensor("valid")]; + tensor var_788_strides_0 = const()[name = tensor("op_788_strides_0"), val = tensor([1, 1])]; + tensor var_788_pad_0 = const()[name = tensor("op_788_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_788_dilations_0 = const()[name = tensor("op_788_dilations_0"), val = tensor([1, 1])]; + tensor var_788_groups_0 = const()[name = tensor("op_788_groups_0"), val = tensor(1)]; + tensor layers_1_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40427584))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42000512))), name = tensor("layers_1_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_788_cast_fp16 = conv(dilations = var_788_dilations_0, groups = var_788_groups_0, pad = var_788_pad_0, pad_type = var_788_pad_type_0, strides = var_788_strides_0, weight = layers_1_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_51_cast_fp16)[name = tensor("op_788_cast_fp16")]; + tensor var_794_pad_type_0 = const()[name = tensor("op_794_pad_type_0"), val = tensor("valid")]; + tensor var_794_strides_0 = const()[name = tensor("op_794_strides_0"), val = tensor([1, 1])]; + tensor var_794_pad_0 = const()[name = tensor("op_794_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_794_dilations_0 = const()[name = tensor("op_794_dilations_0"), val = tensor([1, 1])]; + tensor var_794_groups_0 = const()[name = tensor("op_794_groups_0"), val = tensor(1)]; + tensor layers_1_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42068672))), name = tensor("layers_1_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42000704))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_794_cast_fp16 = conv(dilations = var_794_dilations_0, groups = var_794_groups_0, pad = var_794_pad_0, pad_type = var_794_pad_type_0, strides = var_794_strides_0, weight = layers_1_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_51_cast_fp16)[name = tensor("op_794_cast_fp16")]; + tensor input_53_cast_fp16 = add(x = var_788_cast_fp16, y = var_794_cast_fp16)[name = tensor("input_53_cast_fp16")]; + tensor input_55_split_num_splits_0 = const()[name = tensor("input_55_split_num_splits_0"), val = tensor(2)]; + tensor input_55_split_axis_0 = const()[name = tensor("input_55_split_axis_0"), val = tensor(1)]; + tensor input_55_split_cast_fp16_0, tensor input_55_split_cast_fp16_1 = split(axis = input_55_split_axis_0, num_splits = input_55_split_num_splits_0, x = input_53_cast_fp16)[name = tensor("input_55_split_cast_fp16")]; + tensor input_55_split_1_sigmoid_cast_fp16 = sigmoid(x = input_55_split_cast_fp16_1)[name = tensor("input_55_split_1_sigmoid_cast_fp16")]; + tensor input_55_cast_fp16 = mul(x = input_55_split_cast_fp16_0, y = input_55_split_1_sigmoid_cast_fp16)[name = tensor("input_55_cast_fp16")]; + tensor input_57_pad_type_0 = const()[name = tensor("input_57_pad_type_0"), val = tensor("custom")]; + tensor input_57_pad_0 = const()[name = tensor("input_57_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_57_groups_0 = const()[name = tensor("input_57_groups_0"), val = tensor(1024)]; + tensor input_57_strides_0 = const()[name = tensor("input_57_strides_0"), val = tensor([1, 1])]; + tensor input_57_dilations_0 = const()[name = tensor("input_57_dilations_0"), val = tensor([1, 1])]; + tensor const_270_to_fp16 = const()[name = tensor("const_270_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42330880)))]; + tensor const_271_to_fp16 = const()[name = tensor("const_271_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42349376)))]; + tensor input_59_cast_fp16 = conv(bias = const_271_to_fp16, dilations = input_57_dilations_0, groups = input_57_groups_0, pad = input_57_pad_0, pad_type = input_57_pad_type_0, strides = input_57_strides_0, weight = const_270_to_fp16, x = input_55_cast_fp16)[name = tensor("input_59_cast_fp16")]; + tensor input_61_cast_fp16 = silu(x = input_59_cast_fp16)[name = tensor("input_61_cast_fp16")]; + tensor var_816_pad_type_0 = const()[name = tensor("op_816_pad_type_0"), val = tensor("valid")]; + tensor var_816_strides_0 = const()[name = tensor("op_816_strides_0"), val = tensor([1, 1])]; + tensor var_816_pad_0 = const()[name = tensor("op_816_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_816_dilations_0 = const()[name = tensor("op_816_dilations_0"), val = tensor([1, 1])]; + tensor var_816_groups_0 = const()[name = tensor("op_816_groups_0"), val = tensor(1)]; + tensor layers_1_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42351488))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43137984))), name = tensor("layers_1_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_816_cast_fp16 = conv(dilations = var_816_dilations_0, groups = var_816_groups_0, pad = var_816_pad_0, pad_type = var_816_pad_type_0, strides = var_816_strides_0, weight = layers_1_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_61_cast_fp16)[name = tensor("op_816_cast_fp16")]; + tensor var_822_pad_type_0 = const()[name = tensor("op_822_pad_type_0"), val = tensor("valid")]; + tensor var_822_strides_0 = const()[name = tensor("op_822_strides_0"), val = tensor([1, 1])]; + tensor var_822_pad_0 = const()[name = tensor("op_822_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_822_dilations_0 = const()[name = tensor("op_822_dilations_0"), val = tensor([1, 1])]; + tensor var_822_groups_0 = const()[name = tensor("op_822_groups_0"), val = tensor(1)]; + tensor layers_1_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43173376))), name = tensor("layers_1_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43138176))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_822_cast_fp16 = conv(dilations = var_822_dilations_0, groups = var_822_groups_0, pad = var_822_pad_0, pad_type = var_822_pad_type_0, strides = var_822_strides_0, weight = layers_1_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_61_cast_fp16)[name = tensor("op_822_cast_fp16")]; + tensor x_11_cast_fp16 = add(x = var_816_cast_fp16, y = var_822_cast_fp16)[name = tensor("x_11_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = x_11_cast_fp16)[name = tensor("inputs_17_cast_fp16")]; + tensor out_17_axes_0 = const()[name = tensor("out_17_axes_0"), val = tensor([1])]; + tensor var_833_to_fp16 = const()[name = tensor("op_833_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_833_to_fp16, x = inputs_17_cast_fp16)[name = tensor("out_17_cast_fp16")]; + tensor input_63_gamma_0_to_fp16 = const()[name = tensor("input_63_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43304512)))]; + tensor input_63_beta_0_to_fp16 = const()[name = tensor("input_63_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43306624)))]; + tensor input_63_epsilon_0_to_fp16 = const()[name = tensor("input_63_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_63_cast_fp16 = batch_norm(beta = input_63_beta_0_to_fp16, epsilon = input_63_epsilon_0_to_fp16, gamma = input_63_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_17_cast_fp16)[name = tensor("input_63_cast_fp16")]; + tensor var_853_pad_type_0 = const()[name = tensor("op_853_pad_type_0"), val = tensor("valid")]; + tensor var_853_strides_0 = const()[name = tensor("op_853_strides_0"), val = tensor([1, 1])]; + tensor var_853_pad_0 = const()[name = tensor("op_853_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_853_dilations_0 = const()[name = tensor("op_853_dilations_0"), val = tensor([1, 1])]; + tensor var_853_groups_0 = const()[name = tensor("op_853_groups_0"), val = tensor(1)]; + tensor layers_1_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43308736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46454528))), name = tensor("layers_1_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_853_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_853_dilations_0, groups = var_853_groups_0, pad = var_853_pad_0, pad_type = var_853_pad_type_0, strides = var_853_strides_0, weight = layers_1_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_63_cast_fp16)[name = tensor("op_853_cast_fp16")]; + tensor var_859_pad_type_0 = const()[name = tensor("op_859_pad_type_0"), val = tensor("valid")]; + tensor var_859_strides_0 = const()[name = tensor("op_859_strides_0"), val = tensor([1, 1])]; + tensor var_859_pad_0 = const()[name = tensor("op_859_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_859_dilations_0 = const()[name = tensor("op_859_dilations_0"), val = tensor([1, 1])]; + tensor var_859_groups_0 = const()[name = tensor("op_859_groups_0"), val = tensor(1)]; + tensor layers_1_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46617152))), name = tensor("layers_1_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46454720))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_859_cast_fp16 = conv(dilations = var_859_dilations_0, groups = var_859_groups_0, pad = var_859_pad_0, pad_type = var_859_pad_type_0, strides = var_859_strides_0, weight = layers_1_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_63_cast_fp16)[name = tensor("op_859_cast_fp16")]; + tensor input_65_cast_fp16 = add(x = var_853_cast_fp16, y = var_859_cast_fp16)[name = tensor("input_65_cast_fp16")]; + tensor input_67_cast_fp16 = silu(x = input_65_cast_fp16)[name = tensor("input_67_cast_fp16")]; + tensor var_870_pad_type_0 = const()[name = tensor("op_870_pad_type_0"), val = tensor("valid")]; + tensor var_870_strides_0 = const()[name = tensor("op_870_strides_0"), val = tensor([1, 1])]; + tensor var_870_pad_0 = const()[name = tensor("op_870_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_870_dilations_0 = const()[name = tensor("op_870_dilations_0"), val = tensor([1, 1])]; + tensor var_870_groups_0 = const()[name = tensor("op_870_groups_0"), val = tensor(1)]; + tensor layers_1_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47141504))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50287296))), name = tensor("layers_1_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_870_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_870_dilations_0, groups = var_870_groups_0, pad = var_870_pad_0, pad_type = var_870_pad_type_0, strides = var_870_strides_0, weight = layers_1_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_67_cast_fp16)[name = tensor("op_870_cast_fp16")]; + tensor var_876_pad_type_0 = const()[name = tensor("op_876_pad_type_0"), val = tensor("valid")]; + tensor var_876_strides_0 = const()[name = tensor("op_876_strides_0"), val = tensor([1, 1])]; + tensor var_876_pad_0 = const()[name = tensor("op_876_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_876_dilations_0 = const()[name = tensor("op_876_dilations_0"), val = tensor([1, 1])]; + tensor var_876_groups_0 = const()[name = tensor("op_876_groups_0"), val = tensor(1)]; + tensor layers_1_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50470336))), name = tensor("layers_1_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50287488))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_876_cast_fp16 = conv(dilations = var_876_dilations_0, groups = var_876_groups_0, pad = var_876_pad_0, pad_type = var_876_pad_type_0, strides = var_876_strides_0, weight = layers_1_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_67_cast_fp16)[name = tensor("op_876_cast_fp16")]; + tensor x_13_cast_fp16 = add(x = var_870_cast_fp16, y = var_876_cast_fp16)[name = tensor("x_13_cast_fp16")]; + tensor var_878_to_fp16 = const()[name = tensor("op_878_to_fp16"), val = tensor(0x1p-1)]; + tensor var_879_cast_fp16 = mul(x = x_13_cast_fp16, y = var_878_to_fp16)[name = tensor("op_879_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = var_879_cast_fp16)[name = tensor("inputs_19_cast_fp16")]; + tensor out_19_axes_0 = const()[name = tensor("out_19_axes_0"), val = tensor([1])]; + tensor var_889_to_fp16 = const()[name = tensor("op_889_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_889_to_fp16, x = inputs_19_cast_fp16)[name = tensor("out_19_cast_fp16")]; + tensor inputs_21_gamma_0_to_fp16 = const()[name = tensor("inputs_21_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50994688)))]; + tensor inputs_21_beta_0_to_fp16 = const()[name = tensor("inputs_21_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50996800)))]; + tensor inputs_21_epsilon_0_to_fp16 = const()[name = tensor("inputs_21_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_21_cast_fp16 = batch_norm(beta = inputs_21_beta_0_to_fp16, epsilon = inputs_21_epsilon_0_to_fp16, gamma = inputs_21_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_19_cast_fp16)[name = tensor("inputs_21_cast_fp16")]; + tensor var_903 = const()[name = tensor("op_903"), val = tensor(3)]; + tensor out_21_axes_0 = const()[name = tensor("out_21_axes_0"), val = tensor([1])]; + tensor var_934_to_fp16 = const()[name = tensor("op_934_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_934_to_fp16, x = inputs_21_cast_fp16)[name = tensor("out_21_cast_fp16")]; + tensor input_69_gamma_0_to_fp16 = const()[name = tensor("input_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50998912)))]; + tensor input_69_beta_0_to_fp16 = const()[name = tensor("input_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51001024)))]; + tensor input_69_epsilon_0_to_fp16 = const()[name = tensor("input_69_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_69_cast_fp16 = batch_norm(beta = input_69_beta_0_to_fp16, epsilon = input_69_epsilon_0_to_fp16, gamma = input_69_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_21_cast_fp16)[name = tensor("input_69_cast_fp16")]; + tensor var_954_pad_type_0 = const()[name = tensor("op_954_pad_type_0"), val = tensor("valid")]; + tensor var_954_strides_0 = const()[name = tensor("op_954_strides_0"), val = tensor([1, 1])]; + tensor var_954_pad_0 = const()[name = tensor("op_954_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_954_dilations_0 = const()[name = tensor("op_954_dilations_0"), val = tensor([1, 1])]; + tensor var_954_groups_0 = const()[name = tensor("op_954_groups_0"), val = tensor(1)]; + tensor layers_2_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51003136))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54148928))), name = tensor("layers_2_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_954_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_954_dilations_0, groups = var_954_groups_0, pad = var_954_pad_0, pad_type = var_954_pad_type_0, strides = var_954_strides_0, weight = layers_2_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_69_cast_fp16)[name = tensor("op_954_cast_fp16")]; + tensor var_960_pad_type_0 = const()[name = tensor("op_960_pad_type_0"), val = tensor("valid")]; + tensor var_960_strides_0 = const()[name = tensor("op_960_strides_0"), val = tensor([1, 1])]; + tensor var_960_pad_0 = const()[name = tensor("op_960_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_960_dilations_0 = const()[name = tensor("op_960_dilations_0"), val = tensor([1, 1])]; + tensor var_960_groups_0 = const()[name = tensor("op_960_groups_0"), val = tensor(1)]; + tensor layers_2_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54281536))), name = tensor("layers_2_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54149120))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_960_cast_fp16 = conv(dilations = var_960_dilations_0, groups = var_960_groups_0, pad = var_960_pad_0, pad_type = var_960_pad_type_0, strides = var_960_strides_0, weight = layers_2_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_69_cast_fp16)[name = tensor("op_960_cast_fp16")]; + tensor input_71_cast_fp16 = add(x = var_954_cast_fp16, y = var_960_cast_fp16)[name = tensor("input_71_cast_fp16")]; + tensor input_73_cast_fp16 = silu(x = input_71_cast_fp16)[name = tensor("input_73_cast_fp16")]; + tensor var_971_pad_type_0 = const()[name = tensor("op_971_pad_type_0"), val = tensor("valid")]; + tensor var_971_strides_0 = const()[name = tensor("op_971_strides_0"), val = tensor([1, 1])]; + tensor var_971_pad_0 = const()[name = tensor("op_971_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_971_dilations_0 = const()[name = tensor("op_971_dilations_0"), val = tensor([1, 1])]; + tensor var_971_groups_0 = const()[name = tensor("op_971_groups_0"), val = tensor(1)]; + tensor layers_2_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54805888))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57951680))), name = tensor("layers_2_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_971_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_971_dilations_0, groups = var_971_groups_0, pad = var_971_pad_0, pad_type = var_971_pad_type_0, strides = var_971_strides_0, weight = layers_2_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_73_cast_fp16)[name = tensor("op_971_cast_fp16")]; + tensor var_977_pad_type_0 = const()[name = tensor("op_977_pad_type_0"), val = tensor("valid")]; + tensor var_977_strides_0 = const()[name = tensor("op_977_strides_0"), val = tensor([1, 1])]; + tensor var_977_pad_0 = const()[name = tensor("op_977_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_977_dilations_0 = const()[name = tensor("op_977_dilations_0"), val = tensor([1, 1])]; + tensor var_977_groups_0 = const()[name = tensor("op_977_groups_0"), val = tensor(1)]; + tensor layers_2_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58113088))), name = tensor("layers_2_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57951872))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_977_cast_fp16 = conv(dilations = var_977_dilations_0, groups = var_977_groups_0, pad = var_977_pad_0, pad_type = var_977_pad_type_0, strides = var_977_strides_0, weight = layers_2_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_73_cast_fp16)[name = tensor("op_977_cast_fp16")]; + tensor x_15_cast_fp16 = add(x = var_971_cast_fp16, y = var_977_cast_fp16)[name = tensor("x_15_cast_fp16")]; + tensor var_979_to_fp16 = const()[name = tensor("op_979_to_fp16"), val = tensor(0x1p-1)]; + tensor var_980_cast_fp16 = mul(x = x_15_cast_fp16, y = var_979_to_fp16)[name = tensor("op_980_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = var_980_cast_fp16)[name = tensor("inputs_23_cast_fp16")]; + tensor out_23_axes_0 = const()[name = tensor("out_23_axes_0"), val = tensor([1])]; + tensor var_990_to_fp16 = const()[name = tensor("op_990_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_990_to_fp16, x = inputs_23_cast_fp16)[name = tensor("out_23_cast_fp16")]; + tensor obj_11_gamma_0_to_fp16 = const()[name = tensor("obj_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58637440)))]; + tensor obj_11_beta_0_to_fp16 = const()[name = tensor("obj_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58639552)))]; + tensor obj_11_epsilon_0_to_fp16 = const()[name = tensor("obj_11_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_11_cast_fp16 = batch_norm(beta = obj_11_beta_0_to_fp16, epsilon = obj_11_epsilon_0_to_fp16, gamma = obj_11_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_23_cast_fp16)[name = tensor("obj_11_cast_fp16")]; + tensor var_1015_pad_type_0 = const()[name = tensor("op_1015_pad_type_0"), val = tensor("valid")]; + tensor var_1015_strides_0 = const()[name = tensor("op_1015_strides_0"), val = tensor([1, 1])]; + tensor var_1015_pad_0 = const()[name = tensor("op_1015_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1015_dilations_0 = const()[name = tensor("op_1015_dilations_0"), val = tensor([1, 1])]; + tensor var_1015_groups_0 = const()[name = tensor("op_1015_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58641664))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59428160))), name = tensor("layers_2_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1015_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1015_dilations_0, groups = var_1015_groups_0, pad = var_1015_pad_0, pad_type = var_1015_pad_type_0, strides = var_1015_strides_0, weight = layers_2_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_11_cast_fp16)[name = tensor("op_1015_cast_fp16")]; + tensor var_1021_pad_type_0 = const()[name = tensor("op_1021_pad_type_0"), val = tensor("valid")]; + tensor var_1021_strides_0 = const()[name = tensor("op_1021_strides_0"), val = tensor([1, 1])]; + tensor var_1021_pad_0 = const()[name = tensor("op_1021_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1021_dilations_0 = const()[name = tensor("op_1021_dilations_0"), val = tensor([1, 1])]; + tensor var_1021_groups_0 = const()[name = tensor("op_1021_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59465856))), name = tensor("layers_2_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59428352))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1021_cast_fp16 = conv(dilations = var_1021_dilations_0, groups = var_1021_groups_0, pad = var_1021_pad_0, pad_type = var_1021_pad_type_0, strides = var_1021_strides_0, weight = layers_2_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_11_cast_fp16)[name = tensor("op_1021_cast_fp16")]; + tensor query_9_cast_fp16 = add(x = var_1015_cast_fp16, y = var_1021_cast_fp16)[name = tensor("query_9_cast_fp16")]; + tensor var_1030_pad_type_0 = const()[name = tensor("op_1030_pad_type_0"), val = tensor("valid")]; + tensor var_1030_strides_0 = const()[name = tensor("op_1030_strides_0"), val = tensor([1, 1])]; + tensor var_1030_pad_0 = const()[name = tensor("op_1030_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1030_dilations_0 = const()[name = tensor("op_1030_dilations_0"), val = tensor([1, 1])]; + tensor var_1030_groups_0 = const()[name = tensor("op_1030_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59596992))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60383488))), name = tensor("layers_2_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1030_cast_fp16 = conv(dilations = var_1030_dilations_0, groups = var_1030_groups_0, pad = var_1030_pad_0, pad_type = var_1030_pad_type_0, strides = var_1030_strides_0, weight = layers_2_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_11_cast_fp16)[name = tensor("op_1030_cast_fp16")]; + tensor var_1036_pad_type_0 = const()[name = tensor("op_1036_pad_type_0"), val = tensor("valid")]; + tensor var_1036_strides_0 = const()[name = tensor("op_1036_strides_0"), val = tensor([1, 1])]; + tensor var_1036_pad_0 = const()[name = tensor("op_1036_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1036_dilations_0 = const()[name = tensor("op_1036_dilations_0"), val = tensor([1, 1])]; + tensor var_1036_groups_0 = const()[name = tensor("op_1036_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60424640))), name = tensor("layers_2_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60383680))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1036_cast_fp16 = conv(dilations = var_1036_dilations_0, groups = var_1036_groups_0, pad = var_1036_pad_0, pad_type = var_1036_pad_type_0, strides = var_1036_strides_0, weight = layers_2_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_11_cast_fp16)[name = tensor("op_1036_cast_fp16")]; + tensor key_5_cast_fp16 = add(x = var_1030_cast_fp16, y = var_1036_cast_fp16)[name = tensor("key_5_cast_fp16")]; + tensor var_1046_pad_type_0 = const()[name = tensor("op_1046_pad_type_0"), val = tensor("valid")]; + tensor var_1046_strides_0 = const()[name = tensor("op_1046_strides_0"), val = tensor([1, 1])]; + tensor var_1046_pad_0 = const()[name = tensor("op_1046_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1046_dilations_0 = const()[name = tensor("op_1046_dilations_0"), val = tensor([1, 1])]; + tensor var_1046_groups_0 = const()[name = tensor("op_1046_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60555776))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61342272))), name = tensor("layers_2_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1046_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1046_dilations_0, groups = var_1046_groups_0, pad = var_1046_pad_0, pad_type = var_1046_pad_type_0, strides = var_1046_strides_0, weight = layers_2_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_11_cast_fp16)[name = tensor("op_1046_cast_fp16")]; + tensor var_1052_pad_type_0 = const()[name = tensor("op_1052_pad_type_0"), val = tensor("valid")]; + tensor var_1052_strides_0 = const()[name = tensor("op_1052_strides_0"), val = tensor([1, 1])]; + tensor var_1052_pad_0 = const()[name = tensor("op_1052_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1052_dilations_0 = const()[name = tensor("op_1052_dilations_0"), val = tensor([1, 1])]; + tensor var_1052_groups_0 = const()[name = tensor("op_1052_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61374016))), name = tensor("layers_2_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61342464))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1052_cast_fp16 = conv(dilations = var_1052_dilations_0, groups = var_1052_groups_0, pad = var_1052_pad_0, pad_type = var_1052_pad_type_0, strides = var_1052_strides_0, weight = layers_2_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_11_cast_fp16)[name = tensor("op_1052_cast_fp16")]; + tensor value_5_cast_fp16 = add(x = var_1046_cast_fp16, y = var_1052_cast_fp16)[name = tensor("value_5_cast_fp16")]; + tensor var_1055_to_fp16 = const()[name = tensor("op_1055_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61505152)))]; + tensor query_11_cast_fp16 = add(x = query_9_cast_fp16, y = var_1055_to_fp16)[name = tensor("query_11_cast_fp16")]; + tensor var_1058_to_fp16 = const()[name = tensor("op_1058_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61507264)))]; + tensor q_with_bias_v_5_cast_fp16 = add(x = query_9_cast_fp16, y = var_1058_to_fp16)[name = tensor("q_with_bias_v_5_cast_fp16")]; + tensor var_1068_pad_type_0 = const()[name = tensor("op_1068_pad_type_0"), val = tensor("valid")]; + tensor var_1068_strides_0 = const()[name = tensor("op_1068_strides_0"), val = tensor([1, 1])]; + tensor var_1068_pad_0 = const()[name = tensor("op_1068_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1068_dilations_0 = const()[name = tensor("op_1068_dilations_0"), val = tensor([1, 1])]; + tensor var_1068_groups_0 = const()[name = tensor("op_1068_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61509376))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62295872))), name = tensor("layers_2_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1068_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1068_dilations_0, groups = var_1068_groups_0, pad = var_1068_pad_0, pad_type = var_1068_pad_type_0, strides = var_1068_strides_0, weight = layers_2_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_1068_cast_fp16")]; + tensor var_1074_pad_type_0 = const()[name = tensor("op_1074_pad_type_0"), val = tensor("valid")]; + tensor var_1074_strides_0 = const()[name = tensor("op_1074_strides_0"), val = tensor([1, 1])]; + tensor var_1074_pad_0 = const()[name = tensor("op_1074_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1074_dilations_0 = const()[name = tensor("op_1074_dilations_0"), val = tensor([1, 1])]; + tensor var_1074_groups_0 = const()[name = tensor("op_1074_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62377280))), name = tensor("layers_2_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62296064))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1074_cast_fp16 = conv(dilations = var_1074_dilations_0, groups = var_1074_groups_0, pad = var_1074_pad_0, pad_type = var_1074_pad_type_0, strides = var_1074_strides_0, weight = layers_2_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_1074_cast_fp16")]; + tensor p_5_cast_fp16 = add(x = var_1068_cast_fp16, y = var_1074_cast_fp16)[name = tensor("p_5_cast_fp16")]; + tensor var_1078 = const()[name = tensor("op_1078"), val = tensor([1, 8, 128, 188])]; + tensor var_1079_cast_fp16 = reshape(shape = var_1078, x = q_with_bias_v_5_cast_fp16)[name = tensor("op_1079_cast_fp16")]; + tensor var_1080 = const()[name = tensor("op_1080"), val = tensor([1, 8, 128, -1])]; + tensor var_1081_cast_fp16 = reshape(shape = var_1080, x = p_5_cast_fp16)[name = tensor("op_1081_cast_fp16")]; + tensor matrix_bd_17_transpose_x_0 = const()[name = tensor("matrix_bd_17_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_17_transpose_y_0 = const()[name = tensor("matrix_bd_17_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_17_cast_fp16 = matmul(transpose_x = matrix_bd_17_transpose_x_0, transpose_y = matrix_bd_17_transpose_y_0, x = var_1079_cast_fp16, y = var_1081_cast_fp16)[name = tensor("matrix_bd_17_cast_fp16")]; + tensor matrix_bd_19_pad_0 = const()[name = tensor("matrix_bd_19_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_19_mode_0 = const()[name = tensor("matrix_bd_19_mode_0"), val = tensor("constant")]; + tensor const_32_to_fp16 = const()[name = tensor("const_32_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_19_cast_fp16 = pad(constant_val = const_32_to_fp16, mode = matrix_bd_19_mode_0, pad = matrix_bd_19_pad_0, x = matrix_bd_17_cast_fp16)[name = tensor("matrix_bd_19_cast_fp16")]; + tensor var_1090 = const()[name = tensor("op_1090"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_21_cast_fp16 = reshape(shape = var_1090, x = matrix_bd_19_cast_fp16)[name = tensor("matrix_bd_21_cast_fp16")]; + tensor var_1094_begin_0 = const()[name = tensor("op_1094_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_1094_end_0 = const()[name = tensor("op_1094_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_1094_end_mask_0 = const()[name = tensor("op_1094_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1094_cast_fp16 = slice_by_index(begin = var_1094_begin_0, end = var_1094_end_0, end_mask = var_1094_end_mask_0, x = matrix_bd_21_cast_fp16)[name = tensor("op_1094_cast_fp16")]; + tensor var_1095 = const()[name = tensor("op_1095"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_23_cast_fp16 = reshape(shape = var_1095, x = var_1094_cast_fp16)[name = tensor("matrix_bd_23_cast_fp16")]; + tensor var_1100_begin_0 = const()[name = tensor("op_1100_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1100_end_0 = const()[name = tensor("op_1100_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_1100_end_mask_0 = const()[name = tensor("op_1100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1100_cast_fp16 = slice_by_index(begin = var_1100_begin_0, end = var_1100_end_0, end_mask = var_1100_end_mask_0, x = matrix_bd_23_cast_fp16)[name = tensor("op_1100_cast_fp16")]; + tensor var_1101_to_fp16 = const()[name = tensor("op_1101_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_5_cast_fp16 = mul(x = var_1100_cast_fp16, y = var_1101_to_fp16)[name = tensor("qk_mask_5_cast_fp16")]; + tensor var_1105 = const()[name = tensor("op_1105"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_5_cast_fp16 = reshape(shape = var_1105, x = query_11_cast_fp16)[name = tensor("mh_q_5_cast_fp16")]; + tensor var_1107_to_fp16 = const()[name = tensor("op_1107_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_1108_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_1107_to_fp16)[name = tensor("op_1108_cast_fp16")]; + tensor var_1111 = const()[name = tensor("op_1111"), val = tensor([1, 8, 128, 188])]; + tensor var_1112_cast_fp16 = reshape(shape = var_1111, x = key_5_cast_fp16)[name = tensor("op_1112_cast_fp16")]; + tensor mh_w_9_transpose_x_0 = const()[name = tensor("mh_w_9_transpose_x_0"), val = tensor(true)]; + tensor mh_w_9_transpose_y_0 = const()[name = tensor("mh_w_9_transpose_y_0"), val = tensor(false)]; + tensor mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_1108_cast_fp16, y = var_1112_cast_fp16)[name = tensor("mh_w_9_cast_fp16")]; + tensor mh_w_11_cast_fp16 = add(x = mh_w_9_cast_fp16, y = qk_mask_5_cast_fp16)[name = tensor("mh_w_11_cast_fp16")]; + tensor var_1116_cast_fp16 = softmax(axis = var_903, x = mh_w_11_cast_fp16)[name = tensor("op_1116_cast_fp16")]; + tensor var_1117 = const()[name = tensor("op_1117"), val = tensor([1, 8, 128, 188])]; + tensor var_1118_cast_fp16 = reshape(shape = var_1117, x = value_5_cast_fp16)[name = tensor("op_1118_cast_fp16")]; + tensor attn_5_transpose_x_0 = const()[name = tensor("attn_5_transpose_x_0"), val = tensor(false)]; + tensor attn_5_transpose_y_0 = const()[name = tensor("attn_5_transpose_y_0"), val = tensor(true)]; + tensor attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_1118_cast_fp16, y = var_1116_cast_fp16)[name = tensor("attn_5_cast_fp16")]; + tensor var_1121 = const()[name = tensor("op_1121"), val = tensor([1, 1024, 1, 188])]; + tensor input_75_cast_fp16 = reshape(shape = var_1121, x = attn_5_cast_fp16)[name = tensor("input_75_cast_fp16")]; + tensor var_1131_pad_type_0 = const()[name = tensor("op_1131_pad_type_0"), val = tensor("valid")]; + tensor var_1131_strides_0 = const()[name = tensor("op_1131_strides_0"), val = tensor([1, 1])]; + tensor var_1131_pad_0 = const()[name = tensor("op_1131_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1131_dilations_0 = const()[name = tensor("op_1131_dilations_0"), val = tensor([1, 1])]; + tensor var_1131_groups_0 = const()[name = tensor("op_1131_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62508416))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63294912))), name = tensor("layers_2_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1131_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1131_dilations_0, groups = var_1131_groups_0, pad = var_1131_pad_0, pad_type = var_1131_pad_type_0, strides = var_1131_strides_0, weight = layers_2_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_75_cast_fp16)[name = tensor("op_1131_cast_fp16")]; + tensor var_1137_pad_type_0 = const()[name = tensor("op_1137_pad_type_0"), val = tensor("valid")]; + tensor var_1137_strides_0 = const()[name = tensor("op_1137_strides_0"), val = tensor([1, 1])]; + tensor var_1137_pad_0 = const()[name = tensor("op_1137_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1137_dilations_0 = const()[name = tensor("op_1137_dilations_0"), val = tensor([1, 1])]; + tensor var_1137_groups_0 = const()[name = tensor("op_1137_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63327680))), name = tensor("layers_2_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63295104))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1137_cast_fp16 = conv(dilations = var_1137_dilations_0, groups = var_1137_groups_0, pad = var_1137_pad_0, pad_type = var_1137_pad_type_0, strides = var_1137_strides_0, weight = layers_2_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_75_cast_fp16)[name = tensor("op_1137_cast_fp16")]; + tensor obj_13_cast_fp16 = add(x = var_1131_cast_fp16, y = var_1137_cast_fp16)[name = tensor("obj_13_cast_fp16")]; + tensor inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = obj_13_cast_fp16)[name = tensor("inputs_25_cast_fp16")]; + tensor out_25_axes_0 = const()[name = tensor("out_25_axes_0"), val = tensor([1])]; + tensor var_1148_to_fp16 = const()[name = tensor("op_1148_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_1148_to_fp16, x = inputs_25_cast_fp16)[name = tensor("out_25_cast_fp16")]; + tensor input_77_gamma_0_to_fp16 = const()[name = tensor("input_77_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63458816)))]; + tensor input_77_beta_0_to_fp16 = const()[name = tensor("input_77_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63460928)))]; + tensor input_77_epsilon_0_to_fp16 = const()[name = tensor("input_77_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_77_cast_fp16 = batch_norm(beta = input_77_beta_0_to_fp16, epsilon = input_77_epsilon_0_to_fp16, gamma = input_77_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_25_cast_fp16)[name = tensor("input_77_cast_fp16")]; + tensor var_1169_pad_type_0 = const()[name = tensor("op_1169_pad_type_0"), val = tensor("valid")]; + tensor var_1169_strides_0 = const()[name = tensor("op_1169_strides_0"), val = tensor([1, 1])]; + tensor var_1169_pad_0 = const()[name = tensor("op_1169_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1169_dilations_0 = const()[name = tensor("op_1169_dilations_0"), val = tensor([1, 1])]; + tensor var_1169_groups_0 = const()[name = tensor("op_1169_groups_0"), val = tensor(1)]; + tensor layers_2_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63463040))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65035968))), name = tensor("layers_2_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_1169_cast_fp16 = conv(dilations = var_1169_dilations_0, groups = var_1169_groups_0, pad = var_1169_pad_0, pad_type = var_1169_pad_type_0, strides = var_1169_strides_0, weight = layers_2_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_77_cast_fp16)[name = tensor("op_1169_cast_fp16")]; + tensor var_1175_pad_type_0 = const()[name = tensor("op_1175_pad_type_0"), val = tensor("valid")]; + tensor var_1175_strides_0 = const()[name = tensor("op_1175_strides_0"), val = tensor([1, 1])]; + tensor var_1175_pad_0 = const()[name = tensor("op_1175_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1175_dilations_0 = const()[name = tensor("op_1175_dilations_0"), val = tensor([1, 1])]; + tensor var_1175_groups_0 = const()[name = tensor("op_1175_groups_0"), val = tensor(1)]; + tensor layers_2_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65102464))), name = tensor("layers_2_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65036160))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_1175_cast_fp16 = conv(dilations = var_1175_dilations_0, groups = var_1175_groups_0, pad = var_1175_pad_0, pad_type = var_1175_pad_type_0, strides = var_1175_strides_0, weight = layers_2_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_77_cast_fp16)[name = tensor("op_1175_cast_fp16")]; + tensor input_79_cast_fp16 = add(x = var_1169_cast_fp16, y = var_1175_cast_fp16)[name = tensor("input_79_cast_fp16")]; + tensor input_81_split_num_splits_0 = const()[name = tensor("input_81_split_num_splits_0"), val = tensor(2)]; + tensor input_81_split_axis_0 = const()[name = tensor("input_81_split_axis_0"), val = tensor(1)]; + tensor input_81_split_cast_fp16_0, tensor input_81_split_cast_fp16_1 = split(axis = input_81_split_axis_0, num_splits = input_81_split_num_splits_0, x = input_79_cast_fp16)[name = tensor("input_81_split_cast_fp16")]; + tensor input_81_split_1_sigmoid_cast_fp16 = sigmoid(x = input_81_split_cast_fp16_1)[name = tensor("input_81_split_1_sigmoid_cast_fp16")]; + tensor input_81_cast_fp16 = mul(x = input_81_split_cast_fp16_0, y = input_81_split_1_sigmoid_cast_fp16)[name = tensor("input_81_cast_fp16")]; + tensor input_83_pad_type_0 = const()[name = tensor("input_83_pad_type_0"), val = tensor("custom")]; + tensor input_83_pad_0 = const()[name = tensor("input_83_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_83_groups_0 = const()[name = tensor("input_83_groups_0"), val = tensor(1024)]; + tensor input_83_strides_0 = const()[name = tensor("input_83_strides_0"), val = tensor([1, 1])]; + tensor input_83_dilations_0 = const()[name = tensor("input_83_dilations_0"), val = tensor([1, 1])]; + tensor const_272_to_fp16 = const()[name = tensor("const_272_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65364672)))]; + tensor const_273_to_fp16 = const()[name = tensor("const_273_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65383168)))]; + tensor input_85_cast_fp16 = conv(bias = const_273_to_fp16, dilations = input_83_dilations_0, groups = input_83_groups_0, pad = input_83_pad_0, pad_type = input_83_pad_type_0, strides = input_83_strides_0, weight = const_272_to_fp16, x = input_81_cast_fp16)[name = tensor("input_85_cast_fp16")]; + tensor input_87_cast_fp16 = silu(x = input_85_cast_fp16)[name = tensor("input_87_cast_fp16")]; + tensor var_1197_pad_type_0 = const()[name = tensor("op_1197_pad_type_0"), val = tensor("valid")]; + tensor var_1197_strides_0 = const()[name = tensor("op_1197_strides_0"), val = tensor([1, 1])]; + tensor var_1197_pad_0 = const()[name = tensor("op_1197_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1197_dilations_0 = const()[name = tensor("op_1197_dilations_0"), val = tensor([1, 1])]; + tensor var_1197_groups_0 = const()[name = tensor("op_1197_groups_0"), val = tensor(1)]; + tensor layers_2_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65385280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66171776))), name = tensor("layers_2_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1197_cast_fp16 = conv(dilations = var_1197_dilations_0, groups = var_1197_groups_0, pad = var_1197_pad_0, pad_type = var_1197_pad_type_0, strides = var_1197_strides_0, weight = layers_2_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_87_cast_fp16)[name = tensor("op_1197_cast_fp16")]; + tensor var_1203_pad_type_0 = const()[name = tensor("op_1203_pad_type_0"), val = tensor("valid")]; + tensor var_1203_strides_0 = const()[name = tensor("op_1203_strides_0"), val = tensor([1, 1])]; + tensor var_1203_pad_0 = const()[name = tensor("op_1203_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1203_dilations_0 = const()[name = tensor("op_1203_dilations_0"), val = tensor([1, 1])]; + tensor var_1203_groups_0 = const()[name = tensor("op_1203_groups_0"), val = tensor(1)]; + tensor layers_2_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66205952))), name = tensor("layers_2_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66171968))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1203_cast_fp16 = conv(dilations = var_1203_dilations_0, groups = var_1203_groups_0, pad = var_1203_pad_0, pad_type = var_1203_pad_type_0, strides = var_1203_strides_0, weight = layers_2_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_87_cast_fp16)[name = tensor("op_1203_cast_fp16")]; + tensor x_17_cast_fp16 = add(x = var_1197_cast_fp16, y = var_1203_cast_fp16)[name = tensor("x_17_cast_fp16")]; + tensor inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = x_17_cast_fp16)[name = tensor("inputs_27_cast_fp16")]; + tensor out_27_axes_0 = const()[name = tensor("out_27_axes_0"), val = tensor([1])]; + tensor var_1214_to_fp16 = const()[name = tensor("op_1214_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_1214_to_fp16, x = inputs_27_cast_fp16)[name = tensor("out_27_cast_fp16")]; + tensor input_89_gamma_0_to_fp16 = const()[name = tensor("input_89_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66337088)))]; + tensor input_89_beta_0_to_fp16 = const()[name = tensor("input_89_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66339200)))]; + tensor input_89_epsilon_0_to_fp16 = const()[name = tensor("input_89_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_89_cast_fp16 = batch_norm(beta = input_89_beta_0_to_fp16, epsilon = input_89_epsilon_0_to_fp16, gamma = input_89_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_27_cast_fp16)[name = tensor("input_89_cast_fp16")]; + tensor var_1234_pad_type_0 = const()[name = tensor("op_1234_pad_type_0"), val = tensor("valid")]; + tensor var_1234_strides_0 = const()[name = tensor("op_1234_strides_0"), val = tensor([1, 1])]; + tensor var_1234_pad_0 = const()[name = tensor("op_1234_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1234_dilations_0 = const()[name = tensor("op_1234_dilations_0"), val = tensor([1, 1])]; + tensor var_1234_groups_0 = const()[name = tensor("op_1234_groups_0"), val = tensor(1)]; + tensor layers_2_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66341312))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69487104))), name = tensor("layers_2_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_1234_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_1234_dilations_0, groups = var_1234_groups_0, pad = var_1234_pad_0, pad_type = var_1234_pad_type_0, strides = var_1234_strides_0, weight = layers_2_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_89_cast_fp16)[name = tensor("op_1234_cast_fp16")]; + tensor var_1240_pad_type_0 = const()[name = tensor("op_1240_pad_type_0"), val = tensor("valid")]; + tensor var_1240_strides_0 = const()[name = tensor("op_1240_strides_0"), val = tensor([1, 1])]; + tensor var_1240_pad_0 = const()[name = tensor("op_1240_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1240_dilations_0 = const()[name = tensor("op_1240_dilations_0"), val = tensor([1, 1])]; + tensor var_1240_groups_0 = const()[name = tensor("op_1240_groups_0"), val = tensor(1)]; + tensor layers_2_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69642112))), name = tensor("layers_2_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69487296))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_1240_cast_fp16 = conv(dilations = var_1240_dilations_0, groups = var_1240_groups_0, pad = var_1240_pad_0, pad_type = var_1240_pad_type_0, strides = var_1240_strides_0, weight = layers_2_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_89_cast_fp16)[name = tensor("op_1240_cast_fp16")]; + tensor input_91_cast_fp16 = add(x = var_1234_cast_fp16, y = var_1240_cast_fp16)[name = tensor("input_91_cast_fp16")]; + tensor input_93_cast_fp16 = silu(x = input_91_cast_fp16)[name = tensor("input_93_cast_fp16")]; + tensor var_1251_pad_type_0 = const()[name = tensor("op_1251_pad_type_0"), val = tensor("valid")]; + tensor var_1251_strides_0 = const()[name = tensor("op_1251_strides_0"), val = tensor([1, 1])]; + tensor var_1251_pad_0 = const()[name = tensor("op_1251_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1251_dilations_0 = const()[name = tensor("op_1251_dilations_0"), val = tensor([1, 1])]; + tensor var_1251_groups_0 = const()[name = tensor("op_1251_groups_0"), val = tensor(1)]; + tensor layers_2_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(70166464))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73312256))), name = tensor("layers_2_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_1251_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1251_dilations_0, groups = var_1251_groups_0, pad = var_1251_pad_0, pad_type = var_1251_pad_type_0, strides = var_1251_strides_0, weight = layers_2_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_93_cast_fp16)[name = tensor("op_1251_cast_fp16")]; + tensor var_1257_pad_type_0 = const()[name = tensor("op_1257_pad_type_0"), val = tensor("valid")]; + tensor var_1257_strides_0 = const()[name = tensor("op_1257_strides_0"), val = tensor([1, 1])]; + tensor var_1257_pad_0 = const()[name = tensor("op_1257_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1257_dilations_0 = const()[name = tensor("op_1257_dilations_0"), val = tensor([1, 1])]; + tensor var_1257_groups_0 = const()[name = tensor("op_1257_groups_0"), val = tensor(1)]; + tensor layers_2_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73487872))), name = tensor("layers_2_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73312448))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_1257_cast_fp16 = conv(dilations = var_1257_dilations_0, groups = var_1257_groups_0, pad = var_1257_pad_0, pad_type = var_1257_pad_type_0, strides = var_1257_strides_0, weight = layers_2_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_93_cast_fp16)[name = tensor("op_1257_cast_fp16")]; + tensor x_19_cast_fp16 = add(x = var_1251_cast_fp16, y = var_1257_cast_fp16)[name = tensor("x_19_cast_fp16")]; + tensor var_1259_to_fp16 = const()[name = tensor("op_1259_to_fp16"), val = tensor(0x1p-1)]; + tensor var_1260_cast_fp16 = mul(x = x_19_cast_fp16, y = var_1259_to_fp16)[name = tensor("op_1260_cast_fp16")]; + tensor inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = var_1260_cast_fp16)[name = tensor("inputs_29_cast_fp16")]; + tensor out_29_axes_0 = const()[name = tensor("out_29_axes_0"), val = tensor([1])]; + tensor var_1270_to_fp16 = const()[name = tensor("op_1270_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_1270_to_fp16, x = inputs_29_cast_fp16)[name = tensor("out_29_cast_fp16")]; + tensor inputs_31_gamma_0_to_fp16 = const()[name = tensor("inputs_31_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74012224)))]; + tensor inputs_31_beta_0_to_fp16 = const()[name = tensor("inputs_31_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74014336)))]; + tensor inputs_31_epsilon_0_to_fp16 = const()[name = tensor("inputs_31_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_31_cast_fp16 = batch_norm(beta = inputs_31_beta_0_to_fp16, epsilon = inputs_31_epsilon_0_to_fp16, gamma = inputs_31_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_29_cast_fp16)[name = tensor("inputs_31_cast_fp16")]; + tensor var_1284 = const()[name = tensor("op_1284"), val = tensor(3)]; + tensor out_31_axes_0 = const()[name = tensor("out_31_axes_0"), val = tensor([1])]; + tensor var_1315_to_fp16 = const()[name = tensor("op_1315_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_1315_to_fp16, x = inputs_31_cast_fp16)[name = tensor("out_31_cast_fp16")]; + tensor input_95_gamma_0_to_fp16 = const()[name = tensor("input_95_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74016448)))]; + tensor input_95_beta_0_to_fp16 = const()[name = tensor("input_95_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74018560)))]; + tensor input_95_epsilon_0_to_fp16 = const()[name = tensor("input_95_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_95_cast_fp16 = batch_norm(beta = input_95_beta_0_to_fp16, epsilon = input_95_epsilon_0_to_fp16, gamma = input_95_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_31_cast_fp16)[name = tensor("input_95_cast_fp16")]; + tensor var_1335_pad_type_0 = const()[name = tensor("op_1335_pad_type_0"), val = tensor("valid")]; + tensor var_1335_strides_0 = const()[name = tensor("op_1335_strides_0"), val = tensor([1, 1])]; + tensor var_1335_pad_0 = const()[name = tensor("op_1335_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1335_dilations_0 = const()[name = tensor("op_1335_dilations_0"), val = tensor([1, 1])]; + tensor var_1335_groups_0 = const()[name = tensor("op_1335_groups_0"), val = tensor(1)]; + tensor layers_3_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74020672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77166464))), name = tensor("layers_3_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_1335_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_1335_dilations_0, groups = var_1335_groups_0, pad = var_1335_pad_0, pad_type = var_1335_pad_type_0, strides = var_1335_strides_0, weight = layers_3_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_95_cast_fp16)[name = tensor("op_1335_cast_fp16")]; + tensor var_1341_pad_type_0 = const()[name = tensor("op_1341_pad_type_0"), val = tensor("valid")]; + tensor var_1341_strides_0 = const()[name = tensor("op_1341_strides_0"), val = tensor([1, 1])]; + tensor var_1341_pad_0 = const()[name = tensor("op_1341_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1341_dilations_0 = const()[name = tensor("op_1341_dilations_0"), val = tensor([1, 1])]; + tensor var_1341_groups_0 = const()[name = tensor("op_1341_groups_0"), val = tensor(1)]; + tensor layers_3_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77307264))), name = tensor("layers_3_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77166656))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_1341_cast_fp16 = conv(dilations = var_1341_dilations_0, groups = var_1341_groups_0, pad = var_1341_pad_0, pad_type = var_1341_pad_type_0, strides = var_1341_strides_0, weight = layers_3_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_95_cast_fp16)[name = tensor("op_1341_cast_fp16")]; + tensor input_97_cast_fp16 = add(x = var_1335_cast_fp16, y = var_1341_cast_fp16)[name = tensor("input_97_cast_fp16")]; + tensor input_99_cast_fp16 = silu(x = input_97_cast_fp16)[name = tensor("input_99_cast_fp16")]; + tensor var_1352_pad_type_0 = const()[name = tensor("op_1352_pad_type_0"), val = tensor("valid")]; + tensor var_1352_strides_0 = const()[name = tensor("op_1352_strides_0"), val = tensor([1, 1])]; + tensor var_1352_pad_0 = const()[name = tensor("op_1352_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1352_dilations_0 = const()[name = tensor("op_1352_dilations_0"), val = tensor([1, 1])]; + tensor var_1352_groups_0 = const()[name = tensor("op_1352_groups_0"), val = tensor(1)]; + tensor layers_3_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77831616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80977408))), name = tensor("layers_3_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_1352_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1352_dilations_0, groups = var_1352_groups_0, pad = var_1352_pad_0, pad_type = var_1352_pad_type_0, strides = var_1352_strides_0, weight = layers_3_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_99_cast_fp16)[name = tensor("op_1352_cast_fp16")]; + tensor var_1358_pad_type_0 = const()[name = tensor("op_1358_pad_type_0"), val = tensor("valid")]; + tensor var_1358_strides_0 = const()[name = tensor("op_1358_strides_0"), val = tensor([1, 1])]; + tensor var_1358_pad_0 = const()[name = tensor("op_1358_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1358_dilations_0 = const()[name = tensor("op_1358_dilations_0"), val = tensor([1, 1])]; + tensor var_1358_groups_0 = const()[name = tensor("op_1358_groups_0"), val = tensor(1)]; + tensor layers_3_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81157376))), name = tensor("layers_3_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80977600))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_1358_cast_fp16 = conv(dilations = var_1358_dilations_0, groups = var_1358_groups_0, pad = var_1358_pad_0, pad_type = var_1358_pad_type_0, strides = var_1358_strides_0, weight = layers_3_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_99_cast_fp16)[name = tensor("op_1358_cast_fp16")]; + tensor x_21_cast_fp16 = add(x = var_1352_cast_fp16, y = var_1358_cast_fp16)[name = tensor("x_21_cast_fp16")]; + tensor var_1360_to_fp16 = const()[name = tensor("op_1360_to_fp16"), val = tensor(0x1p-1)]; + tensor var_1361_cast_fp16 = mul(x = x_21_cast_fp16, y = var_1360_to_fp16)[name = tensor("op_1361_cast_fp16")]; + tensor inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = var_1361_cast_fp16)[name = tensor("inputs_33_cast_fp16")]; + tensor out_33_axes_0 = const()[name = tensor("out_33_axes_0"), val = tensor([1])]; + tensor var_1371_to_fp16 = const()[name = tensor("op_1371_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_1371_to_fp16, x = inputs_33_cast_fp16)[name = tensor("out_33_cast_fp16")]; + tensor obj_15_gamma_0_to_fp16 = const()[name = tensor("obj_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81681728)))]; + tensor obj_15_beta_0_to_fp16 = const()[name = tensor("obj_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81683840)))]; + tensor obj_15_epsilon_0_to_fp16 = const()[name = tensor("obj_15_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_15_cast_fp16 = batch_norm(beta = obj_15_beta_0_to_fp16, epsilon = obj_15_epsilon_0_to_fp16, gamma = obj_15_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_33_cast_fp16)[name = tensor("obj_15_cast_fp16")]; + tensor var_1396_pad_type_0 = const()[name = tensor("op_1396_pad_type_0"), val = tensor("valid")]; + tensor var_1396_strides_0 = const()[name = tensor("op_1396_strides_0"), val = tensor([1, 1])]; + tensor var_1396_pad_0 = const()[name = tensor("op_1396_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1396_dilations_0 = const()[name = tensor("op_1396_dilations_0"), val = tensor([1, 1])]; + tensor var_1396_groups_0 = const()[name = tensor("op_1396_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81685952))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82472448))), name = tensor("layers_3_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1396_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1396_dilations_0, groups = var_1396_groups_0, pad = var_1396_pad_0, pad_type = var_1396_pad_type_0, strides = var_1396_strides_0, weight = layers_3_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_15_cast_fp16)[name = tensor("op_1396_cast_fp16")]; + tensor var_1402_pad_type_0 = const()[name = tensor("op_1402_pad_type_0"), val = tensor("valid")]; + tensor var_1402_strides_0 = const()[name = tensor("op_1402_strides_0"), val = tensor([1, 1])]; + tensor var_1402_pad_0 = const()[name = tensor("op_1402_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1402_dilations_0 = const()[name = tensor("op_1402_dilations_0"), val = tensor([1, 1])]; + tensor var_1402_groups_0 = const()[name = tensor("op_1402_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82509504))), name = tensor("layers_3_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82472640))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1402_cast_fp16 = conv(dilations = var_1402_dilations_0, groups = var_1402_groups_0, pad = var_1402_pad_0, pad_type = var_1402_pad_type_0, strides = var_1402_strides_0, weight = layers_3_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_15_cast_fp16)[name = tensor("op_1402_cast_fp16")]; + tensor query_13_cast_fp16 = add(x = var_1396_cast_fp16, y = var_1402_cast_fp16)[name = tensor("query_13_cast_fp16")]; + tensor var_1411_pad_type_0 = const()[name = tensor("op_1411_pad_type_0"), val = tensor("valid")]; + tensor var_1411_strides_0 = const()[name = tensor("op_1411_strides_0"), val = tensor([1, 1])]; + tensor var_1411_pad_0 = const()[name = tensor("op_1411_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1411_dilations_0 = const()[name = tensor("op_1411_dilations_0"), val = tensor([1, 1])]; + tensor var_1411_groups_0 = const()[name = tensor("op_1411_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82640640))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83427136))), name = tensor("layers_3_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1411_cast_fp16 = conv(dilations = var_1411_dilations_0, groups = var_1411_groups_0, pad = var_1411_pad_0, pad_type = var_1411_pad_type_0, strides = var_1411_strides_0, weight = layers_3_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_15_cast_fp16)[name = tensor("op_1411_cast_fp16")]; + tensor var_1417_pad_type_0 = const()[name = tensor("op_1417_pad_type_0"), val = tensor("valid")]; + tensor var_1417_strides_0 = const()[name = tensor("op_1417_strides_0"), val = tensor([1, 1])]; + tensor var_1417_pad_0 = const()[name = tensor("op_1417_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1417_dilations_0 = const()[name = tensor("op_1417_dilations_0"), val = tensor([1, 1])]; + tensor var_1417_groups_0 = const()[name = tensor("op_1417_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83466688))), name = tensor("layers_3_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83427328))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1417_cast_fp16 = conv(dilations = var_1417_dilations_0, groups = var_1417_groups_0, pad = var_1417_pad_0, pad_type = var_1417_pad_type_0, strides = var_1417_strides_0, weight = layers_3_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_15_cast_fp16)[name = tensor("op_1417_cast_fp16")]; + tensor key_7_cast_fp16 = add(x = var_1411_cast_fp16, y = var_1417_cast_fp16)[name = tensor("key_7_cast_fp16")]; + tensor var_1427_pad_type_0 = const()[name = tensor("op_1427_pad_type_0"), val = tensor("valid")]; + tensor var_1427_strides_0 = const()[name = tensor("op_1427_strides_0"), val = tensor([1, 1])]; + tensor var_1427_pad_0 = const()[name = tensor("op_1427_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1427_dilations_0 = const()[name = tensor("op_1427_dilations_0"), val = tensor([1, 1])]; + tensor var_1427_groups_0 = const()[name = tensor("op_1427_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83597824))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84384320))), name = tensor("layers_3_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1427_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1427_dilations_0, groups = var_1427_groups_0, pad = var_1427_pad_0, pad_type = var_1427_pad_type_0, strides = var_1427_strides_0, weight = layers_3_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_15_cast_fp16)[name = tensor("op_1427_cast_fp16")]; + tensor var_1433_pad_type_0 = const()[name = tensor("op_1433_pad_type_0"), val = tensor("valid")]; + tensor var_1433_strides_0 = const()[name = tensor("op_1433_strides_0"), val = tensor([1, 1])]; + tensor var_1433_pad_0 = const()[name = tensor("op_1433_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1433_dilations_0 = const()[name = tensor("op_1433_dilations_0"), val = tensor([1, 1])]; + tensor var_1433_groups_0 = const()[name = tensor("op_1433_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84416896))), name = tensor("layers_3_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84384512))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1433_cast_fp16 = conv(dilations = var_1433_dilations_0, groups = var_1433_groups_0, pad = var_1433_pad_0, pad_type = var_1433_pad_type_0, strides = var_1433_strides_0, weight = layers_3_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_15_cast_fp16)[name = tensor("op_1433_cast_fp16")]; + tensor value_7_cast_fp16 = add(x = var_1427_cast_fp16, y = var_1433_cast_fp16)[name = tensor("value_7_cast_fp16")]; + tensor var_1436_to_fp16 = const()[name = tensor("op_1436_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84548032)))]; + tensor query_15_cast_fp16 = add(x = query_13_cast_fp16, y = var_1436_to_fp16)[name = tensor("query_15_cast_fp16")]; + tensor var_1439_to_fp16 = const()[name = tensor("op_1439_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84550144)))]; + tensor q_with_bias_v_7_cast_fp16 = add(x = query_13_cast_fp16, y = var_1439_to_fp16)[name = tensor("q_with_bias_v_7_cast_fp16")]; + tensor var_1449_pad_type_0 = const()[name = tensor("op_1449_pad_type_0"), val = tensor("valid")]; + tensor var_1449_strides_0 = const()[name = tensor("op_1449_strides_0"), val = tensor([1, 1])]; + tensor var_1449_pad_0 = const()[name = tensor("op_1449_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1449_dilations_0 = const()[name = tensor("op_1449_dilations_0"), val = tensor([1, 1])]; + tensor var_1449_groups_0 = const()[name = tensor("op_1449_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84552256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85338752))), name = tensor("layers_3_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1449_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1449_dilations_0, groups = var_1449_groups_0, pad = var_1449_pad_0, pad_type = var_1449_pad_type_0, strides = var_1449_strides_0, weight = layers_3_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_1449_cast_fp16")]; + tensor var_1455_pad_type_0 = const()[name = tensor("op_1455_pad_type_0"), val = tensor("valid")]; + tensor var_1455_strides_0 = const()[name = tensor("op_1455_strides_0"), val = tensor([1, 1])]; + tensor var_1455_pad_0 = const()[name = tensor("op_1455_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1455_dilations_0 = const()[name = tensor("op_1455_dilations_0"), val = tensor([1, 1])]; + tensor var_1455_groups_0 = const()[name = tensor("op_1455_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85417280))), name = tensor("layers_3_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85338944))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1455_cast_fp16 = conv(dilations = var_1455_dilations_0, groups = var_1455_groups_0, pad = var_1455_pad_0, pad_type = var_1455_pad_type_0, strides = var_1455_strides_0, weight = layers_3_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_1455_cast_fp16")]; + tensor p_7_cast_fp16 = add(x = var_1449_cast_fp16, y = var_1455_cast_fp16)[name = tensor("p_7_cast_fp16")]; + tensor var_1459 = const()[name = tensor("op_1459"), val = tensor([1, 8, 128, 188])]; + tensor var_1460_cast_fp16 = reshape(shape = var_1459, x = q_with_bias_v_7_cast_fp16)[name = tensor("op_1460_cast_fp16")]; + tensor var_1461 = const()[name = tensor("op_1461"), val = tensor([1, 8, 128, -1])]; + tensor var_1462_cast_fp16 = reshape(shape = var_1461, x = p_7_cast_fp16)[name = tensor("op_1462_cast_fp16")]; + tensor matrix_bd_25_transpose_x_0 = const()[name = tensor("matrix_bd_25_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_25_transpose_y_0 = const()[name = tensor("matrix_bd_25_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_25_cast_fp16 = matmul(transpose_x = matrix_bd_25_transpose_x_0, transpose_y = matrix_bd_25_transpose_y_0, x = var_1460_cast_fp16, y = var_1462_cast_fp16)[name = tensor("matrix_bd_25_cast_fp16")]; + tensor matrix_bd_27_pad_0 = const()[name = tensor("matrix_bd_27_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_27_mode_0 = const()[name = tensor("matrix_bd_27_mode_0"), val = tensor("constant")]; + tensor const_43_to_fp16 = const()[name = tensor("const_43_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_27_cast_fp16 = pad(constant_val = const_43_to_fp16, mode = matrix_bd_27_mode_0, pad = matrix_bd_27_pad_0, x = matrix_bd_25_cast_fp16)[name = tensor("matrix_bd_27_cast_fp16")]; + tensor var_1471 = const()[name = tensor("op_1471"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_29_cast_fp16 = reshape(shape = var_1471, x = matrix_bd_27_cast_fp16)[name = tensor("matrix_bd_29_cast_fp16")]; + tensor var_1475_begin_0 = const()[name = tensor("op_1475_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_1475_end_0 = const()[name = tensor("op_1475_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_1475_end_mask_0 = const()[name = tensor("op_1475_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1475_cast_fp16 = slice_by_index(begin = var_1475_begin_0, end = var_1475_end_0, end_mask = var_1475_end_mask_0, x = matrix_bd_29_cast_fp16)[name = tensor("op_1475_cast_fp16")]; + tensor var_1476 = const()[name = tensor("op_1476"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_31_cast_fp16 = reshape(shape = var_1476, x = var_1475_cast_fp16)[name = tensor("matrix_bd_31_cast_fp16")]; + tensor var_1481_begin_0 = const()[name = tensor("op_1481_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1481_end_0 = const()[name = tensor("op_1481_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_1481_end_mask_0 = const()[name = tensor("op_1481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1481_cast_fp16 = slice_by_index(begin = var_1481_begin_0, end = var_1481_end_0, end_mask = var_1481_end_mask_0, x = matrix_bd_31_cast_fp16)[name = tensor("op_1481_cast_fp16")]; + tensor var_1482_to_fp16 = const()[name = tensor("op_1482_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_7_cast_fp16 = mul(x = var_1481_cast_fp16, y = var_1482_to_fp16)[name = tensor("qk_mask_7_cast_fp16")]; + tensor var_1486 = const()[name = tensor("op_1486"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_7_cast_fp16 = reshape(shape = var_1486, x = query_15_cast_fp16)[name = tensor("mh_q_7_cast_fp16")]; + tensor var_1488_to_fp16 = const()[name = tensor("op_1488_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_1489_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_1488_to_fp16)[name = tensor("op_1489_cast_fp16")]; + tensor var_1492 = const()[name = tensor("op_1492"), val = tensor([1, 8, 128, 188])]; + tensor var_1493_cast_fp16 = reshape(shape = var_1492, x = key_7_cast_fp16)[name = tensor("op_1493_cast_fp16")]; + tensor mh_w_13_transpose_x_0 = const()[name = tensor("mh_w_13_transpose_x_0"), val = tensor(true)]; + tensor mh_w_13_transpose_y_0 = const()[name = tensor("mh_w_13_transpose_y_0"), val = tensor(false)]; + tensor mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_1489_cast_fp16, y = var_1493_cast_fp16)[name = tensor("mh_w_13_cast_fp16")]; + tensor mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = qk_mask_7_cast_fp16)[name = tensor("mh_w_15_cast_fp16")]; + tensor var_1497_cast_fp16 = softmax(axis = var_1284, x = mh_w_15_cast_fp16)[name = tensor("op_1497_cast_fp16")]; + tensor var_1498 = const()[name = tensor("op_1498"), val = tensor([1, 8, 128, 188])]; + tensor var_1499_cast_fp16 = reshape(shape = var_1498, x = value_7_cast_fp16)[name = tensor("op_1499_cast_fp16")]; + tensor attn_7_transpose_x_0 = const()[name = tensor("attn_7_transpose_x_0"), val = tensor(false)]; + tensor attn_7_transpose_y_0 = const()[name = tensor("attn_7_transpose_y_0"), val = tensor(true)]; + tensor attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_1499_cast_fp16, y = var_1497_cast_fp16)[name = tensor("attn_7_cast_fp16")]; + tensor var_1502 = const()[name = tensor("op_1502"), val = tensor([1, 1024, 1, 188])]; + tensor input_101_cast_fp16 = reshape(shape = var_1502, x = attn_7_cast_fp16)[name = tensor("input_101_cast_fp16")]; + tensor var_1512_pad_type_0 = const()[name = tensor("op_1512_pad_type_0"), val = tensor("valid")]; + tensor var_1512_strides_0 = const()[name = tensor("op_1512_strides_0"), val = tensor([1, 1])]; + tensor var_1512_pad_0 = const()[name = tensor("op_1512_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1512_dilations_0 = const()[name = tensor("op_1512_dilations_0"), val = tensor([1, 1])]; + tensor var_1512_groups_0 = const()[name = tensor("op_1512_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85548416))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86334912))), name = tensor("layers_3_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1512_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1512_dilations_0, groups = var_1512_groups_0, pad = var_1512_pad_0, pad_type = var_1512_pad_type_0, strides = var_1512_strides_0, weight = layers_3_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_101_cast_fp16)[name = tensor("op_1512_cast_fp16")]; + tensor var_1518_pad_type_0 = const()[name = tensor("op_1518_pad_type_0"), val = tensor("valid")]; + tensor var_1518_strides_0 = const()[name = tensor("op_1518_strides_0"), val = tensor([1, 1])]; + tensor var_1518_pad_0 = const()[name = tensor("op_1518_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1518_dilations_0 = const()[name = tensor("op_1518_dilations_0"), val = tensor([1, 1])]; + tensor var_1518_groups_0 = const()[name = tensor("op_1518_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86368512))), name = tensor("layers_3_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86335104))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1518_cast_fp16 = conv(dilations = var_1518_dilations_0, groups = var_1518_groups_0, pad = var_1518_pad_0, pad_type = var_1518_pad_type_0, strides = var_1518_strides_0, weight = layers_3_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_101_cast_fp16)[name = tensor("op_1518_cast_fp16")]; + tensor obj_17_cast_fp16 = add(x = var_1512_cast_fp16, y = var_1518_cast_fp16)[name = tensor("obj_17_cast_fp16")]; + tensor inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_17_cast_fp16)[name = tensor("inputs_35_cast_fp16")]; + tensor out_35_axes_0 = const()[name = tensor("out_35_axes_0"), val = tensor([1])]; + tensor var_1529_to_fp16 = const()[name = tensor("op_1529_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_1529_to_fp16, x = inputs_35_cast_fp16)[name = tensor("out_35_cast_fp16")]; + tensor input_103_gamma_0_to_fp16 = const()[name = tensor("input_103_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86499648)))]; + tensor input_103_beta_0_to_fp16 = const()[name = tensor("input_103_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86501760)))]; + tensor input_103_epsilon_0_to_fp16 = const()[name = tensor("input_103_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_103_cast_fp16 = batch_norm(beta = input_103_beta_0_to_fp16, epsilon = input_103_epsilon_0_to_fp16, gamma = input_103_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_35_cast_fp16)[name = tensor("input_103_cast_fp16")]; + tensor var_1550_pad_type_0 = const()[name = tensor("op_1550_pad_type_0"), val = tensor("valid")]; + tensor var_1550_strides_0 = const()[name = tensor("op_1550_strides_0"), val = tensor([1, 1])]; + tensor var_1550_pad_0 = const()[name = tensor("op_1550_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1550_dilations_0 = const()[name = tensor("op_1550_dilations_0"), val = tensor([1, 1])]; + tensor var_1550_groups_0 = const()[name = tensor("op_1550_groups_0"), val = tensor(1)]; + tensor layers_3_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86503872))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88076800))), name = tensor("layers_3_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_1550_cast_fp16 = conv(dilations = var_1550_dilations_0, groups = var_1550_groups_0, pad = var_1550_pad_0, pad_type = var_1550_pad_type_0, strides = var_1550_strides_0, weight = layers_3_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_103_cast_fp16)[name = tensor("op_1550_cast_fp16")]; + tensor var_1556_pad_type_0 = const()[name = tensor("op_1556_pad_type_0"), val = tensor("valid")]; + tensor var_1556_strides_0 = const()[name = tensor("op_1556_strides_0"), val = tensor([1, 1])]; + tensor var_1556_pad_0 = const()[name = tensor("op_1556_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1556_dilations_0 = const()[name = tensor("op_1556_dilations_0"), val = tensor([1, 1])]; + tensor var_1556_groups_0 = const()[name = tensor("op_1556_groups_0"), val = tensor(1)]; + tensor layers_3_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88142464))), name = tensor("layers_3_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88076992))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_1556_cast_fp16 = conv(dilations = var_1556_dilations_0, groups = var_1556_groups_0, pad = var_1556_pad_0, pad_type = var_1556_pad_type_0, strides = var_1556_strides_0, weight = layers_3_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_103_cast_fp16)[name = tensor("op_1556_cast_fp16")]; + tensor input_105_cast_fp16 = add(x = var_1550_cast_fp16, y = var_1556_cast_fp16)[name = tensor("input_105_cast_fp16")]; + tensor input_107_split_num_splits_0 = const()[name = tensor("input_107_split_num_splits_0"), val = tensor(2)]; + tensor input_107_split_axis_0 = const()[name = tensor("input_107_split_axis_0"), val = tensor(1)]; + tensor input_107_split_cast_fp16_0, tensor input_107_split_cast_fp16_1 = split(axis = input_107_split_axis_0, num_splits = input_107_split_num_splits_0, x = input_105_cast_fp16)[name = tensor("input_107_split_cast_fp16")]; + tensor input_107_split_1_sigmoid_cast_fp16 = sigmoid(x = input_107_split_cast_fp16_1)[name = tensor("input_107_split_1_sigmoid_cast_fp16")]; + tensor input_107_cast_fp16 = mul(x = input_107_split_cast_fp16_0, y = input_107_split_1_sigmoid_cast_fp16)[name = tensor("input_107_cast_fp16")]; + tensor input_109_pad_type_0 = const()[name = tensor("input_109_pad_type_0"), val = tensor("custom")]; + tensor input_109_pad_0 = const()[name = tensor("input_109_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_109_groups_0 = const()[name = tensor("input_109_groups_0"), val = tensor(1024)]; + tensor input_109_strides_0 = const()[name = tensor("input_109_strides_0"), val = tensor([1, 1])]; + tensor input_109_dilations_0 = const()[name = tensor("input_109_dilations_0"), val = tensor([1, 1])]; + tensor const_274_to_fp16 = const()[name = tensor("const_274_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88404672)))]; + tensor const_275_to_fp16 = const()[name = tensor("const_275_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88423168)))]; + tensor input_111_cast_fp16 = conv(bias = const_275_to_fp16, dilations = input_109_dilations_0, groups = input_109_groups_0, pad = input_109_pad_0, pad_type = input_109_pad_type_0, strides = input_109_strides_0, weight = const_274_to_fp16, x = input_107_cast_fp16)[name = tensor("input_111_cast_fp16")]; + tensor input_113_cast_fp16 = silu(x = input_111_cast_fp16)[name = tensor("input_113_cast_fp16")]; + tensor var_1578_pad_type_0 = const()[name = tensor("op_1578_pad_type_0"), val = tensor("valid")]; + tensor var_1578_strides_0 = const()[name = tensor("op_1578_strides_0"), val = tensor([1, 1])]; + tensor var_1578_pad_0 = const()[name = tensor("op_1578_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1578_dilations_0 = const()[name = tensor("op_1578_dilations_0"), val = tensor([1, 1])]; + tensor var_1578_groups_0 = const()[name = tensor("op_1578_groups_0"), val = tensor(1)]; + tensor layers_3_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88425280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89211776))), name = tensor("layers_3_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1578_cast_fp16 = conv(dilations = var_1578_dilations_0, groups = var_1578_groups_0, pad = var_1578_pad_0, pad_type = var_1578_pad_type_0, strides = var_1578_strides_0, weight = layers_3_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_113_cast_fp16)[name = tensor("op_1578_cast_fp16")]; + tensor var_1584_pad_type_0 = const()[name = tensor("op_1584_pad_type_0"), val = tensor("valid")]; + tensor var_1584_strides_0 = const()[name = tensor("op_1584_strides_0"), val = tensor([1, 1])]; + tensor var_1584_pad_0 = const()[name = tensor("op_1584_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1584_dilations_0 = const()[name = tensor("op_1584_dilations_0"), val = tensor([1, 1])]; + tensor var_1584_groups_0 = const()[name = tensor("op_1584_groups_0"), val = tensor(1)]; + tensor layers_3_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89245696))), name = tensor("layers_3_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89211968))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1584_cast_fp16 = conv(dilations = var_1584_dilations_0, groups = var_1584_groups_0, pad = var_1584_pad_0, pad_type = var_1584_pad_type_0, strides = var_1584_strides_0, weight = layers_3_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_113_cast_fp16)[name = tensor("op_1584_cast_fp16")]; + tensor x_23_cast_fp16 = add(x = var_1578_cast_fp16, y = var_1584_cast_fp16)[name = tensor("x_23_cast_fp16")]; + tensor inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = x_23_cast_fp16)[name = tensor("inputs_37_cast_fp16")]; + tensor out_37_axes_0 = const()[name = tensor("out_37_axes_0"), val = tensor([1])]; + tensor var_1595_to_fp16 = const()[name = tensor("op_1595_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_1595_to_fp16, x = inputs_37_cast_fp16)[name = tensor("out_37_cast_fp16")]; + tensor input_115_gamma_0_to_fp16 = const()[name = tensor("input_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89376832)))]; + tensor input_115_beta_0_to_fp16 = const()[name = tensor("input_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89378944)))]; + tensor input_115_epsilon_0_to_fp16 = const()[name = tensor("input_115_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_115_cast_fp16 = batch_norm(beta = input_115_beta_0_to_fp16, epsilon = input_115_epsilon_0_to_fp16, gamma = input_115_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_37_cast_fp16)[name = tensor("input_115_cast_fp16")]; + tensor var_1615_pad_type_0 = const()[name = tensor("op_1615_pad_type_0"), val = tensor("valid")]; + tensor var_1615_strides_0 = const()[name = tensor("op_1615_strides_0"), val = tensor([1, 1])]; + tensor var_1615_pad_0 = const()[name = tensor("op_1615_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1615_dilations_0 = const()[name = tensor("op_1615_dilations_0"), val = tensor([1, 1])]; + tensor var_1615_groups_0 = const()[name = tensor("op_1615_groups_0"), val = tensor(1)]; + tensor layers_3_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89381056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92526848))), name = tensor("layers_3_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_1615_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_1615_dilations_0, groups = var_1615_groups_0, pad = var_1615_pad_0, pad_type = var_1615_pad_type_0, strides = var_1615_strides_0, weight = layers_3_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_115_cast_fp16)[name = tensor("op_1615_cast_fp16")]; + tensor var_1621_pad_type_0 = const()[name = tensor("op_1621_pad_type_0"), val = tensor("valid")]; + tensor var_1621_strides_0 = const()[name = tensor("op_1621_strides_0"), val = tensor([1, 1])]; + tensor var_1621_pad_0 = const()[name = tensor("op_1621_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1621_dilations_0 = const()[name = tensor("op_1621_dilations_0"), val = tensor([1, 1])]; + tensor var_1621_groups_0 = const()[name = tensor("op_1621_groups_0"), val = tensor(1)]; + tensor layers_3_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92690688))), name = tensor("layers_3_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92527040))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_1621_cast_fp16 = conv(dilations = var_1621_dilations_0, groups = var_1621_groups_0, pad = var_1621_pad_0, pad_type = var_1621_pad_type_0, strides = var_1621_strides_0, weight = layers_3_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_115_cast_fp16)[name = tensor("op_1621_cast_fp16")]; + tensor input_117_cast_fp16 = add(x = var_1615_cast_fp16, y = var_1621_cast_fp16)[name = tensor("input_117_cast_fp16")]; + tensor input_119_cast_fp16 = silu(x = input_117_cast_fp16)[name = tensor("input_119_cast_fp16")]; + tensor var_1632_pad_type_0 = const()[name = tensor("op_1632_pad_type_0"), val = tensor("valid")]; + tensor var_1632_strides_0 = const()[name = tensor("op_1632_strides_0"), val = tensor([1, 1])]; + tensor var_1632_pad_0 = const()[name = tensor("op_1632_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1632_dilations_0 = const()[name = tensor("op_1632_dilations_0"), val = tensor([1, 1])]; + tensor var_1632_groups_0 = const()[name = tensor("op_1632_groups_0"), val = tensor(1)]; + tensor layers_3_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93215040))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96360832))), name = tensor("layers_3_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_1632_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1632_dilations_0, groups = var_1632_groups_0, pad = var_1632_pad_0, pad_type = var_1632_pad_type_0, strides = var_1632_strides_0, weight = layers_3_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_119_cast_fp16)[name = tensor("op_1632_cast_fp16")]; + tensor var_1638_pad_type_0 = const()[name = tensor("op_1638_pad_type_0"), val = tensor("valid")]; + tensor var_1638_strides_0 = const()[name = tensor("op_1638_strides_0"), val = tensor([1, 1])]; + tensor var_1638_pad_0 = const()[name = tensor("op_1638_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1638_dilations_0 = const()[name = tensor("op_1638_dilations_0"), val = tensor([1, 1])]; + tensor var_1638_groups_0 = const()[name = tensor("op_1638_groups_0"), val = tensor(1)]; + tensor layers_3_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96548864))), name = tensor("layers_3_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96361024))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_1638_cast_fp16 = conv(dilations = var_1638_dilations_0, groups = var_1638_groups_0, pad = var_1638_pad_0, pad_type = var_1638_pad_type_0, strides = var_1638_strides_0, weight = layers_3_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_119_cast_fp16)[name = tensor("op_1638_cast_fp16")]; + tensor x_25_cast_fp16 = add(x = var_1632_cast_fp16, y = var_1638_cast_fp16)[name = tensor("x_25_cast_fp16")]; + tensor var_1640_to_fp16 = const()[name = tensor("op_1640_to_fp16"), val = tensor(0x1p-1)]; + tensor var_1641_cast_fp16 = mul(x = x_25_cast_fp16, y = var_1640_to_fp16)[name = tensor("op_1641_cast_fp16")]; + tensor inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = var_1641_cast_fp16)[name = tensor("inputs_39_cast_fp16")]; + tensor out_39_axes_0 = const()[name = tensor("out_39_axes_0"), val = tensor([1])]; + tensor var_1651_to_fp16 = const()[name = tensor("op_1651_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_1651_to_fp16, x = inputs_39_cast_fp16)[name = tensor("out_39_cast_fp16")]; + tensor inputs_41_gamma_0_to_fp16 = const()[name = tensor("inputs_41_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97073216)))]; + tensor inputs_41_beta_0_to_fp16 = const()[name = tensor("inputs_41_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97075328)))]; + tensor inputs_41_epsilon_0_to_fp16 = const()[name = tensor("inputs_41_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_41_cast_fp16 = batch_norm(beta = inputs_41_beta_0_to_fp16, epsilon = inputs_41_epsilon_0_to_fp16, gamma = inputs_41_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_39_cast_fp16)[name = tensor("inputs_41_cast_fp16")]; + tensor var_1665 = const()[name = tensor("op_1665"), val = tensor(3)]; + tensor out_41_axes_0 = const()[name = tensor("out_41_axes_0"), val = tensor([1])]; + tensor var_1696_to_fp16 = const()[name = tensor("op_1696_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_1696_to_fp16, x = inputs_41_cast_fp16)[name = tensor("out_41_cast_fp16")]; + tensor input_121_gamma_0_to_fp16 = const()[name = tensor("input_121_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97077440)))]; + tensor input_121_beta_0_to_fp16 = const()[name = tensor("input_121_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97079552)))]; + tensor input_121_epsilon_0_to_fp16 = const()[name = tensor("input_121_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_121_cast_fp16 = batch_norm(beta = input_121_beta_0_to_fp16, epsilon = input_121_epsilon_0_to_fp16, gamma = input_121_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_41_cast_fp16)[name = tensor("input_121_cast_fp16")]; + tensor var_1716_pad_type_0 = const()[name = tensor("op_1716_pad_type_0"), val = tensor("valid")]; + tensor var_1716_strides_0 = const()[name = tensor("op_1716_strides_0"), val = tensor([1, 1])]; + tensor var_1716_pad_0 = const()[name = tensor("op_1716_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1716_dilations_0 = const()[name = tensor("op_1716_dilations_0"), val = tensor([1, 1])]; + tensor var_1716_groups_0 = const()[name = tensor("op_1716_groups_0"), val = tensor(1)]; + tensor layers_4_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97081664))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100227456))), name = tensor("layers_4_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_1716_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_1716_dilations_0, groups = var_1716_groups_0, pad = var_1716_pad_0, pad_type = var_1716_pad_type_0, strides = var_1716_strides_0, weight = layers_4_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_121_cast_fp16)[name = tensor("op_1716_cast_fp16")]; + tensor var_1722_pad_type_0 = const()[name = tensor("op_1722_pad_type_0"), val = tensor("valid")]; + tensor var_1722_strides_0 = const()[name = tensor("op_1722_strides_0"), val = tensor([1, 1])]; + tensor var_1722_pad_0 = const()[name = tensor("op_1722_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1722_dilations_0 = const()[name = tensor("op_1722_dilations_0"), val = tensor([1, 1])]; + tensor var_1722_groups_0 = const()[name = tensor("op_1722_groups_0"), val = tensor(1)]; + tensor layers_4_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100374080))), name = tensor("layers_4_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100227648))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_1722_cast_fp16 = conv(dilations = var_1722_dilations_0, groups = var_1722_groups_0, pad = var_1722_pad_0, pad_type = var_1722_pad_type_0, strides = var_1722_strides_0, weight = layers_4_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_121_cast_fp16)[name = tensor("op_1722_cast_fp16")]; + tensor input_123_cast_fp16 = add(x = var_1716_cast_fp16, y = var_1722_cast_fp16)[name = tensor("input_123_cast_fp16")]; + tensor input_125_cast_fp16 = silu(x = input_123_cast_fp16)[name = tensor("input_125_cast_fp16")]; + tensor var_1733_pad_type_0 = const()[name = tensor("op_1733_pad_type_0"), val = tensor("valid")]; + tensor var_1733_strides_0 = const()[name = tensor("op_1733_strides_0"), val = tensor([1, 1])]; + tensor var_1733_pad_0 = const()[name = tensor("op_1733_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1733_dilations_0 = const()[name = tensor("op_1733_dilations_0"), val = tensor([1, 1])]; + tensor var_1733_groups_0 = const()[name = tensor("op_1733_groups_0"), val = tensor(1)]; + tensor layers_4_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100898432))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104044224))), name = tensor("layers_4_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_1733_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1733_dilations_0, groups = var_1733_groups_0, pad = var_1733_pad_0, pad_type = var_1733_pad_type_0, strides = var_1733_strides_0, weight = layers_4_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_125_cast_fp16)[name = tensor("op_1733_cast_fp16")]; + tensor var_1739_pad_type_0 = const()[name = tensor("op_1739_pad_type_0"), val = tensor("valid")]; + tensor var_1739_strides_0 = const()[name = tensor("op_1739_strides_0"), val = tensor([1, 1])]; + tensor var_1739_pad_0 = const()[name = tensor("op_1739_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1739_dilations_0 = const()[name = tensor("op_1739_dilations_0"), val = tensor([1, 1])]; + tensor var_1739_groups_0 = const()[name = tensor("op_1739_groups_0"), val = tensor(1)]; + tensor layers_4_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104233088))), name = tensor("layers_4_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104044416))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_1739_cast_fp16 = conv(dilations = var_1739_dilations_0, groups = var_1739_groups_0, pad = var_1739_pad_0, pad_type = var_1739_pad_type_0, strides = var_1739_strides_0, weight = layers_4_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_125_cast_fp16)[name = tensor("op_1739_cast_fp16")]; + tensor x_27_cast_fp16 = add(x = var_1733_cast_fp16, y = var_1739_cast_fp16)[name = tensor("x_27_cast_fp16")]; + tensor var_1741_to_fp16 = const()[name = tensor("op_1741_to_fp16"), val = tensor(0x1p-1)]; + tensor var_1742_cast_fp16 = mul(x = x_27_cast_fp16, y = var_1741_to_fp16)[name = tensor("op_1742_cast_fp16")]; + tensor inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = var_1742_cast_fp16)[name = tensor("inputs_43_cast_fp16")]; + tensor out_43_axes_0 = const()[name = tensor("out_43_axes_0"), val = tensor([1])]; + tensor var_1752_to_fp16 = const()[name = tensor("op_1752_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_1752_to_fp16, x = inputs_43_cast_fp16)[name = tensor("out_43_cast_fp16")]; + tensor obj_19_gamma_0_to_fp16 = const()[name = tensor("obj_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104757440)))]; + tensor obj_19_beta_0_to_fp16 = const()[name = tensor("obj_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104759552)))]; + tensor obj_19_epsilon_0_to_fp16 = const()[name = tensor("obj_19_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_19_cast_fp16 = batch_norm(beta = obj_19_beta_0_to_fp16, epsilon = obj_19_epsilon_0_to_fp16, gamma = obj_19_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_43_cast_fp16)[name = tensor("obj_19_cast_fp16")]; + tensor var_1777_pad_type_0 = const()[name = tensor("op_1777_pad_type_0"), val = tensor("valid")]; + tensor var_1777_strides_0 = const()[name = tensor("op_1777_strides_0"), val = tensor([1, 1])]; + tensor var_1777_pad_0 = const()[name = tensor("op_1777_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1777_dilations_0 = const()[name = tensor("op_1777_dilations_0"), val = tensor([1, 1])]; + tensor var_1777_groups_0 = const()[name = tensor("op_1777_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(104761664))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105548160))), name = tensor("layers_4_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1777_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1777_dilations_0, groups = var_1777_groups_0, pad = var_1777_pad_0, pad_type = var_1777_pad_type_0, strides = var_1777_strides_0, weight = layers_4_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_19_cast_fp16)[name = tensor("op_1777_cast_fp16")]; + tensor var_1783_pad_type_0 = const()[name = tensor("op_1783_pad_type_0"), val = tensor("valid")]; + tensor var_1783_strides_0 = const()[name = tensor("op_1783_strides_0"), val = tensor([1, 1])]; + tensor var_1783_pad_0 = const()[name = tensor("op_1783_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1783_dilations_0 = const()[name = tensor("op_1783_dilations_0"), val = tensor([1, 1])]; + tensor var_1783_groups_0 = const()[name = tensor("op_1783_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105584704))), name = tensor("layers_4_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105548352))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1783_cast_fp16 = conv(dilations = var_1783_dilations_0, groups = var_1783_groups_0, pad = var_1783_pad_0, pad_type = var_1783_pad_type_0, strides = var_1783_strides_0, weight = layers_4_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_19_cast_fp16)[name = tensor("op_1783_cast_fp16")]; + tensor query_17_cast_fp16 = add(x = var_1777_cast_fp16, y = var_1783_cast_fp16)[name = tensor("query_17_cast_fp16")]; + tensor var_1792_pad_type_0 = const()[name = tensor("op_1792_pad_type_0"), val = tensor("valid")]; + tensor var_1792_strides_0 = const()[name = tensor("op_1792_strides_0"), val = tensor([1, 1])]; + tensor var_1792_pad_0 = const()[name = tensor("op_1792_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1792_dilations_0 = const()[name = tensor("op_1792_dilations_0"), val = tensor([1, 1])]; + tensor var_1792_groups_0 = const()[name = tensor("op_1792_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105715840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106502336))), name = tensor("layers_4_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1792_cast_fp16 = conv(dilations = var_1792_dilations_0, groups = var_1792_groups_0, pad = var_1792_pad_0, pad_type = var_1792_pad_type_0, strides = var_1792_strides_0, weight = layers_4_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_19_cast_fp16)[name = tensor("op_1792_cast_fp16")]; + tensor var_1798_pad_type_0 = const()[name = tensor("op_1798_pad_type_0"), val = tensor("valid")]; + tensor var_1798_strides_0 = const()[name = tensor("op_1798_strides_0"), val = tensor([1, 1])]; + tensor var_1798_pad_0 = const()[name = tensor("op_1798_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1798_dilations_0 = const()[name = tensor("op_1798_dilations_0"), val = tensor([1, 1])]; + tensor var_1798_groups_0 = const()[name = tensor("op_1798_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106538752))), name = tensor("layers_4_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106502528))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1798_cast_fp16 = conv(dilations = var_1798_dilations_0, groups = var_1798_groups_0, pad = var_1798_pad_0, pad_type = var_1798_pad_type_0, strides = var_1798_strides_0, weight = layers_4_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_19_cast_fp16)[name = tensor("op_1798_cast_fp16")]; + tensor key_9_cast_fp16 = add(x = var_1792_cast_fp16, y = var_1798_cast_fp16)[name = tensor("key_9_cast_fp16")]; + tensor var_1808_pad_type_0 = const()[name = tensor("op_1808_pad_type_0"), val = tensor("valid")]; + tensor var_1808_strides_0 = const()[name = tensor("op_1808_strides_0"), val = tensor([1, 1])]; + tensor var_1808_pad_0 = const()[name = tensor("op_1808_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1808_dilations_0 = const()[name = tensor("op_1808_dilations_0"), val = tensor([1, 1])]; + tensor var_1808_groups_0 = const()[name = tensor("op_1808_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106669888))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107456384))), name = tensor("layers_4_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1808_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1808_dilations_0, groups = var_1808_groups_0, pad = var_1808_pad_0, pad_type = var_1808_pad_type_0, strides = var_1808_strides_0, weight = layers_4_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_19_cast_fp16)[name = tensor("op_1808_cast_fp16")]; + tensor var_1814_pad_type_0 = const()[name = tensor("op_1814_pad_type_0"), val = tensor("valid")]; + tensor var_1814_strides_0 = const()[name = tensor("op_1814_strides_0"), val = tensor([1, 1])]; + tensor var_1814_pad_0 = const()[name = tensor("op_1814_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1814_dilations_0 = const()[name = tensor("op_1814_dilations_0"), val = tensor([1, 1])]; + tensor var_1814_groups_0 = const()[name = tensor("op_1814_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107489984))), name = tensor("layers_4_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107456576))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1814_cast_fp16 = conv(dilations = var_1814_dilations_0, groups = var_1814_groups_0, pad = var_1814_pad_0, pad_type = var_1814_pad_type_0, strides = var_1814_strides_0, weight = layers_4_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_19_cast_fp16)[name = tensor("op_1814_cast_fp16")]; + tensor value_9_cast_fp16 = add(x = var_1808_cast_fp16, y = var_1814_cast_fp16)[name = tensor("value_9_cast_fp16")]; + tensor var_1817_to_fp16 = const()[name = tensor("op_1817_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107621120)))]; + tensor query_19_cast_fp16 = add(x = query_17_cast_fp16, y = var_1817_to_fp16)[name = tensor("query_19_cast_fp16")]; + tensor var_1820_to_fp16 = const()[name = tensor("op_1820_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107623232)))]; + tensor q_with_bias_v_9_cast_fp16 = add(x = query_17_cast_fp16, y = var_1820_to_fp16)[name = tensor("q_with_bias_v_9_cast_fp16")]; + tensor var_1830_pad_type_0 = const()[name = tensor("op_1830_pad_type_0"), val = tensor("valid")]; + tensor var_1830_strides_0 = const()[name = tensor("op_1830_strides_0"), val = tensor([1, 1])]; + tensor var_1830_pad_0 = const()[name = tensor("op_1830_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1830_dilations_0 = const()[name = tensor("op_1830_dilations_0"), val = tensor([1, 1])]; + tensor var_1830_groups_0 = const()[name = tensor("op_1830_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107625344))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108411840))), name = tensor("layers_4_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1830_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1830_dilations_0, groups = var_1830_groups_0, pad = var_1830_pad_0, pad_type = var_1830_pad_type_0, strides = var_1830_strides_0, weight = layers_4_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_1830_cast_fp16")]; + tensor var_1836_pad_type_0 = const()[name = tensor("op_1836_pad_type_0"), val = tensor("valid")]; + tensor var_1836_strides_0 = const()[name = tensor("op_1836_strides_0"), val = tensor([1, 1])]; + tensor var_1836_pad_0 = const()[name = tensor("op_1836_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1836_dilations_0 = const()[name = tensor("op_1836_dilations_0"), val = tensor([1, 1])]; + tensor var_1836_groups_0 = const()[name = tensor("op_1836_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108489280))), name = tensor("layers_4_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108412032))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1836_cast_fp16 = conv(dilations = var_1836_dilations_0, groups = var_1836_groups_0, pad = var_1836_pad_0, pad_type = var_1836_pad_type_0, strides = var_1836_strides_0, weight = layers_4_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_1836_cast_fp16")]; + tensor p_9_cast_fp16 = add(x = var_1830_cast_fp16, y = var_1836_cast_fp16)[name = tensor("p_9_cast_fp16")]; + tensor var_1840 = const()[name = tensor("op_1840"), val = tensor([1, 8, 128, 188])]; + tensor var_1841_cast_fp16 = reshape(shape = var_1840, x = q_with_bias_v_9_cast_fp16)[name = tensor("op_1841_cast_fp16")]; + tensor var_1842 = const()[name = tensor("op_1842"), val = tensor([1, 8, 128, -1])]; + tensor var_1843_cast_fp16 = reshape(shape = var_1842, x = p_9_cast_fp16)[name = tensor("op_1843_cast_fp16")]; + tensor matrix_bd_33_transpose_x_0 = const()[name = tensor("matrix_bd_33_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_33_transpose_y_0 = const()[name = tensor("matrix_bd_33_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_33_cast_fp16 = matmul(transpose_x = matrix_bd_33_transpose_x_0, transpose_y = matrix_bd_33_transpose_y_0, x = var_1841_cast_fp16, y = var_1843_cast_fp16)[name = tensor("matrix_bd_33_cast_fp16")]; + tensor matrix_bd_35_pad_0 = const()[name = tensor("matrix_bd_35_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_35_mode_0 = const()[name = tensor("matrix_bd_35_mode_0"), val = tensor("constant")]; + tensor const_54_to_fp16 = const()[name = tensor("const_54_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_35_cast_fp16 = pad(constant_val = const_54_to_fp16, mode = matrix_bd_35_mode_0, pad = matrix_bd_35_pad_0, x = matrix_bd_33_cast_fp16)[name = tensor("matrix_bd_35_cast_fp16")]; + tensor var_1852 = const()[name = tensor("op_1852"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_37_cast_fp16 = reshape(shape = var_1852, x = matrix_bd_35_cast_fp16)[name = tensor("matrix_bd_37_cast_fp16")]; + tensor var_1856_begin_0 = const()[name = tensor("op_1856_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_1856_end_0 = const()[name = tensor("op_1856_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_1856_end_mask_0 = const()[name = tensor("op_1856_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1856_cast_fp16 = slice_by_index(begin = var_1856_begin_0, end = var_1856_end_0, end_mask = var_1856_end_mask_0, x = matrix_bd_37_cast_fp16)[name = tensor("op_1856_cast_fp16")]; + tensor var_1857 = const()[name = tensor("op_1857"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_39_cast_fp16 = reshape(shape = var_1857, x = var_1856_cast_fp16)[name = tensor("matrix_bd_39_cast_fp16")]; + tensor var_1862_begin_0 = const()[name = tensor("op_1862_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1862_end_0 = const()[name = tensor("op_1862_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_1862_end_mask_0 = const()[name = tensor("op_1862_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1862_cast_fp16 = slice_by_index(begin = var_1862_begin_0, end = var_1862_end_0, end_mask = var_1862_end_mask_0, x = matrix_bd_39_cast_fp16)[name = tensor("op_1862_cast_fp16")]; + tensor var_1863_to_fp16 = const()[name = tensor("op_1863_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_9_cast_fp16 = mul(x = var_1862_cast_fp16, y = var_1863_to_fp16)[name = tensor("qk_mask_9_cast_fp16")]; + tensor var_1867 = const()[name = tensor("op_1867"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_9_cast_fp16 = reshape(shape = var_1867, x = query_19_cast_fp16)[name = tensor("mh_q_9_cast_fp16")]; + tensor var_1869_to_fp16 = const()[name = tensor("op_1869_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_1870_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_1869_to_fp16)[name = tensor("op_1870_cast_fp16")]; + tensor var_1873 = const()[name = tensor("op_1873"), val = tensor([1, 8, 128, 188])]; + tensor var_1874_cast_fp16 = reshape(shape = var_1873, x = key_9_cast_fp16)[name = tensor("op_1874_cast_fp16")]; + tensor mh_w_17_transpose_x_0 = const()[name = tensor("mh_w_17_transpose_x_0"), val = tensor(true)]; + tensor mh_w_17_transpose_y_0 = const()[name = tensor("mh_w_17_transpose_y_0"), val = tensor(false)]; + tensor mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_1870_cast_fp16, y = var_1874_cast_fp16)[name = tensor("mh_w_17_cast_fp16")]; + tensor mh_w_19_cast_fp16 = add(x = mh_w_17_cast_fp16, y = qk_mask_9_cast_fp16)[name = tensor("mh_w_19_cast_fp16")]; + tensor var_1878_cast_fp16 = softmax(axis = var_1665, x = mh_w_19_cast_fp16)[name = tensor("op_1878_cast_fp16")]; + tensor var_1879 = const()[name = tensor("op_1879"), val = tensor([1, 8, 128, 188])]; + tensor var_1880_cast_fp16 = reshape(shape = var_1879, x = value_9_cast_fp16)[name = tensor("op_1880_cast_fp16")]; + tensor attn_9_transpose_x_0 = const()[name = tensor("attn_9_transpose_x_0"), val = tensor(false)]; + tensor attn_9_transpose_y_0 = const()[name = tensor("attn_9_transpose_y_0"), val = tensor(true)]; + tensor attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_1880_cast_fp16, y = var_1878_cast_fp16)[name = tensor("attn_9_cast_fp16")]; + tensor var_1883 = const()[name = tensor("op_1883"), val = tensor([1, 1024, 1, 188])]; + tensor input_127_cast_fp16 = reshape(shape = var_1883, x = attn_9_cast_fp16)[name = tensor("input_127_cast_fp16")]; + tensor var_1893_pad_type_0 = const()[name = tensor("op_1893_pad_type_0"), val = tensor("valid")]; + tensor var_1893_strides_0 = const()[name = tensor("op_1893_strides_0"), val = tensor([1, 1])]; + tensor var_1893_pad_0 = const()[name = tensor("op_1893_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1893_dilations_0 = const()[name = tensor("op_1893_dilations_0"), val = tensor([1, 1])]; + tensor var_1893_groups_0 = const()[name = tensor("op_1893_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108620416))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109406912))), name = tensor("layers_4_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1893_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1893_dilations_0, groups = var_1893_groups_0, pad = var_1893_pad_0, pad_type = var_1893_pad_type_0, strides = var_1893_strides_0, weight = layers_4_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_127_cast_fp16)[name = tensor("op_1893_cast_fp16")]; + tensor var_1899_pad_type_0 = const()[name = tensor("op_1899_pad_type_0"), val = tensor("valid")]; + tensor var_1899_strides_0 = const()[name = tensor("op_1899_strides_0"), val = tensor([1, 1])]; + tensor var_1899_pad_0 = const()[name = tensor("op_1899_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1899_dilations_0 = const()[name = tensor("op_1899_dilations_0"), val = tensor([1, 1])]; + tensor var_1899_groups_0 = const()[name = tensor("op_1899_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109438656))), name = tensor("layers_4_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109407104))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1899_cast_fp16 = conv(dilations = var_1899_dilations_0, groups = var_1899_groups_0, pad = var_1899_pad_0, pad_type = var_1899_pad_type_0, strides = var_1899_strides_0, weight = layers_4_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_127_cast_fp16)[name = tensor("op_1899_cast_fp16")]; + tensor obj_21_cast_fp16 = add(x = var_1893_cast_fp16, y = var_1899_cast_fp16)[name = tensor("obj_21_cast_fp16")]; + tensor inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = obj_21_cast_fp16)[name = tensor("inputs_45_cast_fp16")]; + tensor out_45_axes_0 = const()[name = tensor("out_45_axes_0"), val = tensor([1])]; + tensor var_1910_to_fp16 = const()[name = tensor("op_1910_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_1910_to_fp16, x = inputs_45_cast_fp16)[name = tensor("out_45_cast_fp16")]; + tensor input_129_gamma_0_to_fp16 = const()[name = tensor("input_129_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109569792)))]; + tensor input_129_beta_0_to_fp16 = const()[name = tensor("input_129_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109571904)))]; + tensor input_129_epsilon_0_to_fp16 = const()[name = tensor("input_129_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_129_cast_fp16 = batch_norm(beta = input_129_beta_0_to_fp16, epsilon = input_129_epsilon_0_to_fp16, gamma = input_129_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_45_cast_fp16)[name = tensor("input_129_cast_fp16")]; + tensor var_1931_pad_type_0 = const()[name = tensor("op_1931_pad_type_0"), val = tensor("valid")]; + tensor var_1931_strides_0 = const()[name = tensor("op_1931_strides_0"), val = tensor([1, 1])]; + tensor var_1931_pad_0 = const()[name = tensor("op_1931_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1931_dilations_0 = const()[name = tensor("op_1931_dilations_0"), val = tensor([1, 1])]; + tensor var_1931_groups_0 = const()[name = tensor("op_1931_groups_0"), val = tensor(1)]; + tensor layers_4_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(109574016))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111146944))), name = tensor("layers_4_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_1931_cast_fp16 = conv(dilations = var_1931_dilations_0, groups = var_1931_groups_0, pad = var_1931_pad_0, pad_type = var_1931_pad_type_0, strides = var_1931_strides_0, weight = layers_4_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_129_cast_fp16)[name = tensor("op_1931_cast_fp16")]; + tensor var_1937_pad_type_0 = const()[name = tensor("op_1937_pad_type_0"), val = tensor("valid")]; + tensor var_1937_strides_0 = const()[name = tensor("op_1937_strides_0"), val = tensor([1, 1])]; + tensor var_1937_pad_0 = const()[name = tensor("op_1937_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1937_dilations_0 = const()[name = tensor("op_1937_dilations_0"), val = tensor([1, 1])]; + tensor var_1937_groups_0 = const()[name = tensor("op_1937_groups_0"), val = tensor(1)]; + tensor layers_4_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111212352))), name = tensor("layers_4_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111147136))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_1937_cast_fp16 = conv(dilations = var_1937_dilations_0, groups = var_1937_groups_0, pad = var_1937_pad_0, pad_type = var_1937_pad_type_0, strides = var_1937_strides_0, weight = layers_4_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_129_cast_fp16)[name = tensor("op_1937_cast_fp16")]; + tensor input_131_cast_fp16 = add(x = var_1931_cast_fp16, y = var_1937_cast_fp16)[name = tensor("input_131_cast_fp16")]; + tensor input_133_split_num_splits_0 = const()[name = tensor("input_133_split_num_splits_0"), val = tensor(2)]; + tensor input_133_split_axis_0 = const()[name = tensor("input_133_split_axis_0"), val = tensor(1)]; + tensor input_133_split_cast_fp16_0, tensor input_133_split_cast_fp16_1 = split(axis = input_133_split_axis_0, num_splits = input_133_split_num_splits_0, x = input_131_cast_fp16)[name = tensor("input_133_split_cast_fp16")]; + tensor input_133_split_1_sigmoid_cast_fp16 = sigmoid(x = input_133_split_cast_fp16_1)[name = tensor("input_133_split_1_sigmoid_cast_fp16")]; + tensor input_133_cast_fp16 = mul(x = input_133_split_cast_fp16_0, y = input_133_split_1_sigmoid_cast_fp16)[name = tensor("input_133_cast_fp16")]; + tensor input_135_pad_type_0 = const()[name = tensor("input_135_pad_type_0"), val = tensor("custom")]; + tensor input_135_pad_0 = const()[name = tensor("input_135_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_135_groups_0 = const()[name = tensor("input_135_groups_0"), val = tensor(1024)]; + tensor input_135_strides_0 = const()[name = tensor("input_135_strides_0"), val = tensor([1, 1])]; + tensor input_135_dilations_0 = const()[name = tensor("input_135_dilations_0"), val = tensor([1, 1])]; + tensor const_276_to_fp16 = const()[name = tensor("const_276_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111474560)))]; + tensor const_277_to_fp16 = const()[name = tensor("const_277_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111493056)))]; + tensor input_137_cast_fp16 = conv(bias = const_277_to_fp16, dilations = input_135_dilations_0, groups = input_135_groups_0, pad = input_135_pad_0, pad_type = input_135_pad_type_0, strides = input_135_strides_0, weight = const_276_to_fp16, x = input_133_cast_fp16)[name = tensor("input_137_cast_fp16")]; + tensor input_139_cast_fp16 = silu(x = input_137_cast_fp16)[name = tensor("input_139_cast_fp16")]; + tensor var_1959_pad_type_0 = const()[name = tensor("op_1959_pad_type_0"), val = tensor("valid")]; + tensor var_1959_strides_0 = const()[name = tensor("op_1959_strides_0"), val = tensor([1, 1])]; + tensor var_1959_pad_0 = const()[name = tensor("op_1959_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1959_dilations_0 = const()[name = tensor("op_1959_dilations_0"), val = tensor([1, 1])]; + tensor var_1959_groups_0 = const()[name = tensor("op_1959_groups_0"), val = tensor(1)]; + tensor layers_4_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(111495168))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112281664))), name = tensor("layers_4_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1959_cast_fp16 = conv(dilations = var_1959_dilations_0, groups = var_1959_groups_0, pad = var_1959_pad_0, pad_type = var_1959_pad_type_0, strides = var_1959_strides_0, weight = layers_4_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_139_cast_fp16)[name = tensor("op_1959_cast_fp16")]; + tensor var_1965_pad_type_0 = const()[name = tensor("op_1965_pad_type_0"), val = tensor("valid")]; + tensor var_1965_strides_0 = const()[name = tensor("op_1965_strides_0"), val = tensor([1, 1])]; + tensor var_1965_pad_0 = const()[name = tensor("op_1965_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1965_dilations_0 = const()[name = tensor("op_1965_dilations_0"), val = tensor([1, 1])]; + tensor var_1965_groups_0 = const()[name = tensor("op_1965_groups_0"), val = tensor(1)]; + tensor layers_4_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112314624))), name = tensor("layers_4_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112281856))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_1965_cast_fp16 = conv(dilations = var_1965_dilations_0, groups = var_1965_groups_0, pad = var_1965_pad_0, pad_type = var_1965_pad_type_0, strides = var_1965_strides_0, weight = layers_4_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_139_cast_fp16)[name = tensor("op_1965_cast_fp16")]; + tensor x_29_cast_fp16 = add(x = var_1959_cast_fp16, y = var_1965_cast_fp16)[name = tensor("x_29_cast_fp16")]; + tensor inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = x_29_cast_fp16)[name = tensor("inputs_47_cast_fp16")]; + tensor out_47_axes_0 = const()[name = tensor("out_47_axes_0"), val = tensor([1])]; + tensor var_1976_to_fp16 = const()[name = tensor("op_1976_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_1976_to_fp16, x = inputs_47_cast_fp16)[name = tensor("out_47_cast_fp16")]; + tensor input_141_gamma_0_to_fp16 = const()[name = tensor("input_141_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112445760)))]; + tensor input_141_beta_0_to_fp16 = const()[name = tensor("input_141_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112447872)))]; + tensor input_141_epsilon_0_to_fp16 = const()[name = tensor("input_141_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_141_cast_fp16 = batch_norm(beta = input_141_beta_0_to_fp16, epsilon = input_141_epsilon_0_to_fp16, gamma = input_141_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_47_cast_fp16)[name = tensor("input_141_cast_fp16")]; + tensor var_1996_pad_type_0 = const()[name = tensor("op_1996_pad_type_0"), val = tensor("valid")]; + tensor var_1996_strides_0 = const()[name = tensor("op_1996_strides_0"), val = tensor([1, 1])]; + tensor var_1996_pad_0 = const()[name = tensor("op_1996_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1996_dilations_0 = const()[name = tensor("op_1996_dilations_0"), val = tensor([1, 1])]; + tensor var_1996_groups_0 = const()[name = tensor("op_1996_groups_0"), val = tensor(1)]; + tensor layers_4_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112449984))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(115595776))), name = tensor("layers_4_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_1996_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_1996_dilations_0, groups = var_1996_groups_0, pad = var_1996_pad_0, pad_type = var_1996_pad_type_0, strides = var_1996_strides_0, weight = layers_4_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_141_cast_fp16)[name = tensor("op_1996_cast_fp16")]; + tensor var_2002_pad_type_0 = const()[name = tensor("op_2002_pad_type_0"), val = tensor("valid")]; + tensor var_2002_strides_0 = const()[name = tensor("op_2002_strides_0"), val = tensor([1, 1])]; + tensor var_2002_pad_0 = const()[name = tensor("op_2002_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2002_dilations_0 = const()[name = tensor("op_2002_dilations_0"), val = tensor([1, 1])]; + tensor var_2002_groups_0 = const()[name = tensor("op_2002_groups_0"), val = tensor(1)]; + tensor layers_4_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(115761984))), name = tensor("layers_4_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(115595968))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_2002_cast_fp16 = conv(dilations = var_2002_dilations_0, groups = var_2002_groups_0, pad = var_2002_pad_0, pad_type = var_2002_pad_type_0, strides = var_2002_strides_0, weight = layers_4_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_141_cast_fp16)[name = tensor("op_2002_cast_fp16")]; + tensor input_143_cast_fp16 = add(x = var_1996_cast_fp16, y = var_2002_cast_fp16)[name = tensor("input_143_cast_fp16")]; + tensor input_145_cast_fp16 = silu(x = input_143_cast_fp16)[name = tensor("input_145_cast_fp16")]; + tensor var_2013_pad_type_0 = const()[name = tensor("op_2013_pad_type_0"), val = tensor("valid")]; + tensor var_2013_strides_0 = const()[name = tensor("op_2013_strides_0"), val = tensor([1, 1])]; + tensor var_2013_pad_0 = const()[name = tensor("op_2013_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2013_dilations_0 = const()[name = tensor("op_2013_dilations_0"), val = tensor([1, 1])]; + tensor var_2013_groups_0 = const()[name = tensor("op_2013_groups_0"), val = tensor(1)]; + tensor layers_4_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116286336))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119432128))), name = tensor("layers_4_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_2013_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2013_dilations_0, groups = var_2013_groups_0, pad = var_2013_pad_0, pad_type = var_2013_pad_type_0, strides = var_2013_strides_0, weight = layers_4_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_145_cast_fp16)[name = tensor("op_2013_cast_fp16")]; + tensor var_2019_pad_type_0 = const()[name = tensor("op_2019_pad_type_0"), val = tensor("valid")]; + tensor var_2019_strides_0 = const()[name = tensor("op_2019_strides_0"), val = tensor([1, 1])]; + tensor var_2019_pad_0 = const()[name = tensor("op_2019_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2019_dilations_0 = const()[name = tensor("op_2019_dilations_0"), val = tensor([1, 1])]; + tensor var_2019_groups_0 = const()[name = tensor("op_2019_groups_0"), val = tensor(1)]; + tensor layers_4_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119620864))), name = tensor("layers_4_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119432320))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_2019_cast_fp16 = conv(dilations = var_2019_dilations_0, groups = var_2019_groups_0, pad = var_2019_pad_0, pad_type = var_2019_pad_type_0, strides = var_2019_strides_0, weight = layers_4_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_145_cast_fp16)[name = tensor("op_2019_cast_fp16")]; + tensor x_31_cast_fp16 = add(x = var_2013_cast_fp16, y = var_2019_cast_fp16)[name = tensor("x_31_cast_fp16")]; + tensor var_2021_to_fp16 = const()[name = tensor("op_2021_to_fp16"), val = tensor(0x1p-1)]; + tensor var_2022_cast_fp16 = mul(x = x_31_cast_fp16, y = var_2021_to_fp16)[name = tensor("op_2022_cast_fp16")]; + tensor inputs_49_cast_fp16 = add(x = inputs_47_cast_fp16, y = var_2022_cast_fp16)[name = tensor("inputs_49_cast_fp16")]; + tensor out_49_axes_0 = const()[name = tensor("out_49_axes_0"), val = tensor([1])]; + tensor var_2032_to_fp16 = const()[name = tensor("op_2032_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_49_cast_fp16 = layer_norm(axes = out_49_axes_0, epsilon = var_2032_to_fp16, x = inputs_49_cast_fp16)[name = tensor("out_49_cast_fp16")]; + tensor inputs_51_gamma_0_to_fp16 = const()[name = tensor("inputs_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120145216)))]; + tensor inputs_51_beta_0_to_fp16 = const()[name = tensor("inputs_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120147328)))]; + tensor inputs_51_epsilon_0_to_fp16 = const()[name = tensor("inputs_51_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_51_cast_fp16 = batch_norm(beta = inputs_51_beta_0_to_fp16, epsilon = inputs_51_epsilon_0_to_fp16, gamma = inputs_51_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_49_cast_fp16)[name = tensor("inputs_51_cast_fp16")]; + tensor var_2046 = const()[name = tensor("op_2046"), val = tensor(3)]; + tensor out_51_axes_0 = const()[name = tensor("out_51_axes_0"), val = tensor([1])]; + tensor var_2077_to_fp16 = const()[name = tensor("op_2077_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_51_cast_fp16 = layer_norm(axes = out_51_axes_0, epsilon = var_2077_to_fp16, x = inputs_51_cast_fp16)[name = tensor("out_51_cast_fp16")]; + tensor input_147_gamma_0_to_fp16 = const()[name = tensor("input_147_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120149440)))]; + tensor input_147_beta_0_to_fp16 = const()[name = tensor("input_147_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120151552)))]; + tensor input_147_epsilon_0_to_fp16 = const()[name = tensor("input_147_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_147_cast_fp16 = batch_norm(beta = input_147_beta_0_to_fp16, epsilon = input_147_epsilon_0_to_fp16, gamma = input_147_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_51_cast_fp16)[name = tensor("input_147_cast_fp16")]; + tensor var_2097_pad_type_0 = const()[name = tensor("op_2097_pad_type_0"), val = tensor("valid")]; + tensor var_2097_strides_0 = const()[name = tensor("op_2097_strides_0"), val = tensor([1, 1])]; + tensor var_2097_pad_0 = const()[name = tensor("op_2097_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2097_dilations_0 = const()[name = tensor("op_2097_dilations_0"), val = tensor([1, 1])]; + tensor var_2097_groups_0 = const()[name = tensor("op_2097_groups_0"), val = tensor(1)]; + tensor layers_5_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120153664))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123299456))), name = tensor("layers_5_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_2097_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_2097_dilations_0, groups = var_2097_groups_0, pad = var_2097_pad_0, pad_type = var_2097_pad_type_0, strides = var_2097_strides_0, weight = layers_5_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_147_cast_fp16)[name = tensor("op_2097_cast_fp16")]; + tensor var_2103_pad_type_0 = const()[name = tensor("op_2103_pad_type_0"), val = tensor("valid")]; + tensor var_2103_strides_0 = const()[name = tensor("op_2103_strides_0"), val = tensor([1, 1])]; + tensor var_2103_pad_0 = const()[name = tensor("op_2103_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2103_dilations_0 = const()[name = tensor("op_2103_dilations_0"), val = tensor([1, 1])]; + tensor var_2103_groups_0 = const()[name = tensor("op_2103_groups_0"), val = tensor(1)]; + tensor layers_5_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123439488))), name = tensor("layers_5_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123299648))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_2103_cast_fp16 = conv(dilations = var_2103_dilations_0, groups = var_2103_groups_0, pad = var_2103_pad_0, pad_type = var_2103_pad_type_0, strides = var_2103_strides_0, weight = layers_5_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_147_cast_fp16)[name = tensor("op_2103_cast_fp16")]; + tensor input_149_cast_fp16 = add(x = var_2097_cast_fp16, y = var_2103_cast_fp16)[name = tensor("input_149_cast_fp16")]; + tensor input_151_cast_fp16 = silu(x = input_149_cast_fp16)[name = tensor("input_151_cast_fp16")]; + tensor var_2114_pad_type_0 = const()[name = tensor("op_2114_pad_type_0"), val = tensor("valid")]; + tensor var_2114_strides_0 = const()[name = tensor("op_2114_strides_0"), val = tensor([1, 1])]; + tensor var_2114_pad_0 = const()[name = tensor("op_2114_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2114_dilations_0 = const()[name = tensor("op_2114_dilations_0"), val = tensor([1, 1])]; + tensor var_2114_groups_0 = const()[name = tensor("op_2114_groups_0"), val = tensor(1)]; + tensor layers_5_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123963840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127109632))), name = tensor("layers_5_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_2114_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2114_dilations_0, groups = var_2114_groups_0, pad = var_2114_pad_0, pad_type = var_2114_pad_type_0, strides = var_2114_strides_0, weight = layers_5_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_151_cast_fp16)[name = tensor("op_2114_cast_fp16")]; + tensor var_2120_pad_type_0 = const()[name = tensor("op_2120_pad_type_0"), val = tensor("valid")]; + tensor var_2120_strides_0 = const()[name = tensor("op_2120_strides_0"), val = tensor([1, 1])]; + tensor var_2120_pad_0 = const()[name = tensor("op_2120_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2120_dilations_0 = const()[name = tensor("op_2120_dilations_0"), val = tensor([1, 1])]; + tensor var_2120_groups_0 = const()[name = tensor("op_2120_groups_0"), val = tensor(1)]; + tensor layers_5_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127291072))), name = tensor("layers_5_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127109824))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_2120_cast_fp16 = conv(dilations = var_2120_dilations_0, groups = var_2120_groups_0, pad = var_2120_pad_0, pad_type = var_2120_pad_type_0, strides = var_2120_strides_0, weight = layers_5_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_151_cast_fp16)[name = tensor("op_2120_cast_fp16")]; + tensor x_33_cast_fp16 = add(x = var_2114_cast_fp16, y = var_2120_cast_fp16)[name = tensor("x_33_cast_fp16")]; + tensor var_2122_to_fp16 = const()[name = tensor("op_2122_to_fp16"), val = tensor(0x1p-1)]; + tensor var_2123_cast_fp16 = mul(x = x_33_cast_fp16, y = var_2122_to_fp16)[name = tensor("op_2123_cast_fp16")]; + tensor inputs_53_cast_fp16 = add(x = inputs_51_cast_fp16, y = var_2123_cast_fp16)[name = tensor("inputs_53_cast_fp16")]; + tensor out_53_axes_0 = const()[name = tensor("out_53_axes_0"), val = tensor([1])]; + tensor var_2133_to_fp16 = const()[name = tensor("op_2133_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_53_cast_fp16 = layer_norm(axes = out_53_axes_0, epsilon = var_2133_to_fp16, x = inputs_53_cast_fp16)[name = tensor("out_53_cast_fp16")]; + tensor obj_23_gamma_0_to_fp16 = const()[name = tensor("obj_23_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127815424)))]; + tensor obj_23_beta_0_to_fp16 = const()[name = tensor("obj_23_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127817536)))]; + tensor obj_23_epsilon_0_to_fp16 = const()[name = tensor("obj_23_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_23_cast_fp16 = batch_norm(beta = obj_23_beta_0_to_fp16, epsilon = obj_23_epsilon_0_to_fp16, gamma = obj_23_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_53_cast_fp16)[name = tensor("obj_23_cast_fp16")]; + tensor var_2158_pad_type_0 = const()[name = tensor("op_2158_pad_type_0"), val = tensor("valid")]; + tensor var_2158_strides_0 = const()[name = tensor("op_2158_strides_0"), val = tensor([1, 1])]; + tensor var_2158_pad_0 = const()[name = tensor("op_2158_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2158_dilations_0 = const()[name = tensor("op_2158_dilations_0"), val = tensor([1, 1])]; + tensor var_2158_groups_0 = const()[name = tensor("op_2158_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127819648))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(128606144))), name = tensor("layers_5_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2158_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2158_dilations_0, groups = var_2158_groups_0, pad = var_2158_pad_0, pad_type = var_2158_pad_type_0, strides = var_2158_strides_0, weight = layers_5_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_23_cast_fp16)[name = tensor("op_2158_cast_fp16")]; + tensor var_2164_pad_type_0 = const()[name = tensor("op_2164_pad_type_0"), val = tensor("valid")]; + tensor var_2164_strides_0 = const()[name = tensor("op_2164_strides_0"), val = tensor([1, 1])]; + tensor var_2164_pad_0 = const()[name = tensor("op_2164_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2164_dilations_0 = const()[name = tensor("op_2164_dilations_0"), val = tensor([1, 1])]; + tensor var_2164_groups_0 = const()[name = tensor("op_2164_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(128641600))), name = tensor("layers_5_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(128606336))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2164_cast_fp16 = conv(dilations = var_2164_dilations_0, groups = var_2164_groups_0, pad = var_2164_pad_0, pad_type = var_2164_pad_type_0, strides = var_2164_strides_0, weight = layers_5_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_23_cast_fp16)[name = tensor("op_2164_cast_fp16")]; + tensor query_21_cast_fp16 = add(x = var_2158_cast_fp16, y = var_2164_cast_fp16)[name = tensor("query_21_cast_fp16")]; + tensor var_2173_pad_type_0 = const()[name = tensor("op_2173_pad_type_0"), val = tensor("valid")]; + tensor var_2173_strides_0 = const()[name = tensor("op_2173_strides_0"), val = tensor([1, 1])]; + tensor var_2173_pad_0 = const()[name = tensor("op_2173_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2173_dilations_0 = const()[name = tensor("op_2173_dilations_0"), val = tensor([1, 1])]; + tensor var_2173_groups_0 = const()[name = tensor("op_2173_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(128772736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129559232))), name = tensor("layers_5_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2173_cast_fp16 = conv(dilations = var_2173_dilations_0, groups = var_2173_groups_0, pad = var_2173_pad_0, pad_type = var_2173_pad_type_0, strides = var_2173_strides_0, weight = layers_5_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_23_cast_fp16)[name = tensor("op_2173_cast_fp16")]; + tensor var_2179_pad_type_0 = const()[name = tensor("op_2179_pad_type_0"), val = tensor("valid")]; + tensor var_2179_strides_0 = const()[name = tensor("op_2179_strides_0"), val = tensor([1, 1])]; + tensor var_2179_pad_0 = const()[name = tensor("op_2179_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2179_dilations_0 = const()[name = tensor("op_2179_dilations_0"), val = tensor([1, 1])]; + tensor var_2179_groups_0 = const()[name = tensor("op_2179_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129600192))), name = tensor("layers_5_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129559424))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2179_cast_fp16 = conv(dilations = var_2179_dilations_0, groups = var_2179_groups_0, pad = var_2179_pad_0, pad_type = var_2179_pad_type_0, strides = var_2179_strides_0, weight = layers_5_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_23_cast_fp16)[name = tensor("op_2179_cast_fp16")]; + tensor key_11_cast_fp16 = add(x = var_2173_cast_fp16, y = var_2179_cast_fp16)[name = tensor("key_11_cast_fp16")]; + tensor var_2189_pad_type_0 = const()[name = tensor("op_2189_pad_type_0"), val = tensor("valid")]; + tensor var_2189_strides_0 = const()[name = tensor("op_2189_strides_0"), val = tensor([1, 1])]; + tensor var_2189_pad_0 = const()[name = tensor("op_2189_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2189_dilations_0 = const()[name = tensor("op_2189_dilations_0"), val = tensor([1, 1])]; + tensor var_2189_groups_0 = const()[name = tensor("op_2189_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129731328))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130517824))), name = tensor("layers_5_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2189_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2189_dilations_0, groups = var_2189_groups_0, pad = var_2189_pad_0, pad_type = var_2189_pad_type_0, strides = var_2189_strides_0, weight = layers_5_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_23_cast_fp16)[name = tensor("op_2189_cast_fp16")]; + tensor var_2195_pad_type_0 = const()[name = tensor("op_2195_pad_type_0"), val = tensor("valid")]; + tensor var_2195_strides_0 = const()[name = tensor("op_2195_strides_0"), val = tensor([1, 1])]; + tensor var_2195_pad_0 = const()[name = tensor("op_2195_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2195_dilations_0 = const()[name = tensor("op_2195_dilations_0"), val = tensor([1, 1])]; + tensor var_2195_groups_0 = const()[name = tensor("op_2195_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130551232))), name = tensor("layers_5_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130518016))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2195_cast_fp16 = conv(dilations = var_2195_dilations_0, groups = var_2195_groups_0, pad = var_2195_pad_0, pad_type = var_2195_pad_type_0, strides = var_2195_strides_0, weight = layers_5_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_23_cast_fp16)[name = tensor("op_2195_cast_fp16")]; + tensor value_11_cast_fp16 = add(x = var_2189_cast_fp16, y = var_2195_cast_fp16)[name = tensor("value_11_cast_fp16")]; + tensor var_2198_to_fp16 = const()[name = tensor("op_2198_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130682368)))]; + tensor query_23_cast_fp16 = add(x = query_21_cast_fp16, y = var_2198_to_fp16)[name = tensor("query_23_cast_fp16")]; + tensor var_2201_to_fp16 = const()[name = tensor("op_2201_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130684480)))]; + tensor q_with_bias_v_11_cast_fp16 = add(x = query_21_cast_fp16, y = var_2201_to_fp16)[name = tensor("q_with_bias_v_11_cast_fp16")]; + tensor var_2211_pad_type_0 = const()[name = tensor("op_2211_pad_type_0"), val = tensor("valid")]; + tensor var_2211_strides_0 = const()[name = tensor("op_2211_strides_0"), val = tensor([1, 1])]; + tensor var_2211_pad_0 = const()[name = tensor("op_2211_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2211_dilations_0 = const()[name = tensor("op_2211_dilations_0"), val = tensor([1, 1])]; + tensor var_2211_groups_0 = const()[name = tensor("op_2211_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(130686592))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131473088))), name = tensor("layers_5_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2211_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2211_dilations_0, groups = var_2211_groups_0, pad = var_2211_pad_0, pad_type = var_2211_pad_type_0, strides = var_2211_strides_0, weight = layers_5_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_2211_cast_fp16")]; + tensor var_2217_pad_type_0 = const()[name = tensor("op_2217_pad_type_0"), val = tensor("valid")]; + tensor var_2217_strides_0 = const()[name = tensor("op_2217_strides_0"), val = tensor([1, 1])]; + tensor var_2217_pad_0 = const()[name = tensor("op_2217_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2217_dilations_0 = const()[name = tensor("op_2217_dilations_0"), val = tensor([1, 1])]; + tensor var_2217_groups_0 = const()[name = tensor("op_2217_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131550336))), name = tensor("layers_5_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131473280))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2217_cast_fp16 = conv(dilations = var_2217_dilations_0, groups = var_2217_groups_0, pad = var_2217_pad_0, pad_type = var_2217_pad_type_0, strides = var_2217_strides_0, weight = layers_5_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_2217_cast_fp16")]; + tensor p_11_cast_fp16 = add(x = var_2211_cast_fp16, y = var_2217_cast_fp16)[name = tensor("p_11_cast_fp16")]; + tensor var_2221 = const()[name = tensor("op_2221"), val = tensor([1, 8, 128, 188])]; + tensor var_2222_cast_fp16 = reshape(shape = var_2221, x = q_with_bias_v_11_cast_fp16)[name = tensor("op_2222_cast_fp16")]; + tensor var_2223 = const()[name = tensor("op_2223"), val = tensor([1, 8, 128, -1])]; + tensor var_2224_cast_fp16 = reshape(shape = var_2223, x = p_11_cast_fp16)[name = tensor("op_2224_cast_fp16")]; + tensor matrix_bd_41_transpose_x_0 = const()[name = tensor("matrix_bd_41_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_41_transpose_y_0 = const()[name = tensor("matrix_bd_41_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_41_cast_fp16 = matmul(transpose_x = matrix_bd_41_transpose_x_0, transpose_y = matrix_bd_41_transpose_y_0, x = var_2222_cast_fp16, y = var_2224_cast_fp16)[name = tensor("matrix_bd_41_cast_fp16")]; + tensor matrix_bd_43_pad_0 = const()[name = tensor("matrix_bd_43_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_43_mode_0 = const()[name = tensor("matrix_bd_43_mode_0"), val = tensor("constant")]; + tensor const_65_to_fp16 = const()[name = tensor("const_65_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_43_cast_fp16 = pad(constant_val = const_65_to_fp16, mode = matrix_bd_43_mode_0, pad = matrix_bd_43_pad_0, x = matrix_bd_41_cast_fp16)[name = tensor("matrix_bd_43_cast_fp16")]; + tensor var_2233 = const()[name = tensor("op_2233"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_45_cast_fp16 = reshape(shape = var_2233, x = matrix_bd_43_cast_fp16)[name = tensor("matrix_bd_45_cast_fp16")]; + tensor var_2237_begin_0 = const()[name = tensor("op_2237_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_2237_end_0 = const()[name = tensor("op_2237_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_2237_end_mask_0 = const()[name = tensor("op_2237_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2237_cast_fp16 = slice_by_index(begin = var_2237_begin_0, end = var_2237_end_0, end_mask = var_2237_end_mask_0, x = matrix_bd_45_cast_fp16)[name = tensor("op_2237_cast_fp16")]; + tensor var_2238 = const()[name = tensor("op_2238"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_47_cast_fp16 = reshape(shape = var_2238, x = var_2237_cast_fp16)[name = tensor("matrix_bd_47_cast_fp16")]; + tensor var_2243_begin_0 = const()[name = tensor("op_2243_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2243_end_0 = const()[name = tensor("op_2243_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_2243_end_mask_0 = const()[name = tensor("op_2243_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2243_cast_fp16 = slice_by_index(begin = var_2243_begin_0, end = var_2243_end_0, end_mask = var_2243_end_mask_0, x = matrix_bd_47_cast_fp16)[name = tensor("op_2243_cast_fp16")]; + tensor var_2244_to_fp16 = const()[name = tensor("op_2244_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_11_cast_fp16 = mul(x = var_2243_cast_fp16, y = var_2244_to_fp16)[name = tensor("qk_mask_11_cast_fp16")]; + tensor var_2248 = const()[name = tensor("op_2248"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_11_cast_fp16 = reshape(shape = var_2248, x = query_23_cast_fp16)[name = tensor("mh_q_11_cast_fp16")]; + tensor var_2250_to_fp16 = const()[name = tensor("op_2250_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_2251_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_2250_to_fp16)[name = tensor("op_2251_cast_fp16")]; + tensor var_2254 = const()[name = tensor("op_2254"), val = tensor([1, 8, 128, 188])]; + tensor var_2255_cast_fp16 = reshape(shape = var_2254, x = key_11_cast_fp16)[name = tensor("op_2255_cast_fp16")]; + tensor mh_w_21_transpose_x_0 = const()[name = tensor("mh_w_21_transpose_x_0"), val = tensor(true)]; + tensor mh_w_21_transpose_y_0 = const()[name = tensor("mh_w_21_transpose_y_0"), val = tensor(false)]; + tensor mh_w_21_cast_fp16 = matmul(transpose_x = mh_w_21_transpose_x_0, transpose_y = mh_w_21_transpose_y_0, x = var_2251_cast_fp16, y = var_2255_cast_fp16)[name = tensor("mh_w_21_cast_fp16")]; + tensor mh_w_23_cast_fp16 = add(x = mh_w_21_cast_fp16, y = qk_mask_11_cast_fp16)[name = tensor("mh_w_23_cast_fp16")]; + tensor var_2259_cast_fp16 = softmax(axis = var_2046, x = mh_w_23_cast_fp16)[name = tensor("op_2259_cast_fp16")]; + tensor var_2260 = const()[name = tensor("op_2260"), val = tensor([1, 8, 128, 188])]; + tensor var_2261_cast_fp16 = reshape(shape = var_2260, x = value_11_cast_fp16)[name = tensor("op_2261_cast_fp16")]; + tensor attn_11_transpose_x_0 = const()[name = tensor("attn_11_transpose_x_0"), val = tensor(false)]; + tensor attn_11_transpose_y_0 = const()[name = tensor("attn_11_transpose_y_0"), val = tensor(true)]; + tensor attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_2261_cast_fp16, y = var_2259_cast_fp16)[name = tensor("attn_11_cast_fp16")]; + tensor var_2264 = const()[name = tensor("op_2264"), val = tensor([1, 1024, 1, 188])]; + tensor input_153_cast_fp16 = reshape(shape = var_2264, x = attn_11_cast_fp16)[name = tensor("input_153_cast_fp16")]; + tensor var_2274_pad_type_0 = const()[name = tensor("op_2274_pad_type_0"), val = tensor("valid")]; + tensor var_2274_strides_0 = const()[name = tensor("op_2274_strides_0"), val = tensor([1, 1])]; + tensor var_2274_pad_0 = const()[name = tensor("op_2274_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2274_dilations_0 = const()[name = tensor("op_2274_dilations_0"), val = tensor([1, 1])]; + tensor var_2274_groups_0 = const()[name = tensor("op_2274_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(131681472))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132467968))), name = tensor("layers_5_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2274_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2274_dilations_0, groups = var_2274_groups_0, pad = var_2274_pad_0, pad_type = var_2274_pad_type_0, strides = var_2274_strides_0, weight = layers_5_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_153_cast_fp16)[name = tensor("op_2274_cast_fp16")]; + tensor var_2280_pad_type_0 = const()[name = tensor("op_2280_pad_type_0"), val = tensor("valid")]; + tensor var_2280_strides_0 = const()[name = tensor("op_2280_strides_0"), val = tensor([1, 1])]; + tensor var_2280_pad_0 = const()[name = tensor("op_2280_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2280_dilations_0 = const()[name = tensor("op_2280_dilations_0"), val = tensor([1, 1])]; + tensor var_2280_groups_0 = const()[name = tensor("op_2280_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132499712))), name = tensor("layers_5_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132468160))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2280_cast_fp16 = conv(dilations = var_2280_dilations_0, groups = var_2280_groups_0, pad = var_2280_pad_0, pad_type = var_2280_pad_type_0, strides = var_2280_strides_0, weight = layers_5_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_153_cast_fp16)[name = tensor("op_2280_cast_fp16")]; + tensor obj_25_cast_fp16 = add(x = var_2274_cast_fp16, y = var_2280_cast_fp16)[name = tensor("obj_25_cast_fp16")]; + tensor inputs_55_cast_fp16 = add(x = inputs_53_cast_fp16, y = obj_25_cast_fp16)[name = tensor("inputs_55_cast_fp16")]; + tensor out_55_axes_0 = const()[name = tensor("out_55_axes_0"), val = tensor([1])]; + tensor var_2291_to_fp16 = const()[name = tensor("op_2291_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_55_cast_fp16 = layer_norm(axes = out_55_axes_0, epsilon = var_2291_to_fp16, x = inputs_55_cast_fp16)[name = tensor("out_55_cast_fp16")]; + tensor input_155_gamma_0_to_fp16 = const()[name = tensor("input_155_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132630848)))]; + tensor input_155_beta_0_to_fp16 = const()[name = tensor("input_155_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132632960)))]; + tensor input_155_epsilon_0_to_fp16 = const()[name = tensor("input_155_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_155_cast_fp16 = batch_norm(beta = input_155_beta_0_to_fp16, epsilon = input_155_epsilon_0_to_fp16, gamma = input_155_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_55_cast_fp16)[name = tensor("input_155_cast_fp16")]; + tensor var_2312_pad_type_0 = const()[name = tensor("op_2312_pad_type_0"), val = tensor("valid")]; + tensor var_2312_strides_0 = const()[name = tensor("op_2312_strides_0"), val = tensor([1, 1])]; + tensor var_2312_pad_0 = const()[name = tensor("op_2312_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2312_dilations_0 = const()[name = tensor("op_2312_dilations_0"), val = tensor([1, 1])]; + tensor var_2312_groups_0 = const()[name = tensor("op_2312_groups_0"), val = tensor(1)]; + tensor layers_5_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132635072))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134208000))), name = tensor("layers_5_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_2312_cast_fp16 = conv(dilations = var_2312_dilations_0, groups = var_2312_groups_0, pad = var_2312_pad_0, pad_type = var_2312_pad_type_0, strides = var_2312_strides_0, weight = layers_5_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_155_cast_fp16)[name = tensor("op_2312_cast_fp16")]; + tensor var_2318_pad_type_0 = const()[name = tensor("op_2318_pad_type_0"), val = tensor("valid")]; + tensor var_2318_strides_0 = const()[name = tensor("op_2318_strides_0"), val = tensor([1, 1])]; + tensor var_2318_pad_0 = const()[name = tensor("op_2318_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2318_dilations_0 = const()[name = tensor("op_2318_dilations_0"), val = tensor([1, 1])]; + tensor var_2318_groups_0 = const()[name = tensor("op_2318_groups_0"), val = tensor(1)]; + tensor layers_5_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134272128))), name = tensor("layers_5_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134208192))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_2318_cast_fp16 = conv(dilations = var_2318_dilations_0, groups = var_2318_groups_0, pad = var_2318_pad_0, pad_type = var_2318_pad_type_0, strides = var_2318_strides_0, weight = layers_5_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_155_cast_fp16)[name = tensor("op_2318_cast_fp16")]; + tensor input_157_cast_fp16 = add(x = var_2312_cast_fp16, y = var_2318_cast_fp16)[name = tensor("input_157_cast_fp16")]; + tensor input_159_split_num_splits_0 = const()[name = tensor("input_159_split_num_splits_0"), val = tensor(2)]; + tensor input_159_split_axis_0 = const()[name = tensor("input_159_split_axis_0"), val = tensor(1)]; + tensor input_159_split_cast_fp16_0, tensor input_159_split_cast_fp16_1 = split(axis = input_159_split_axis_0, num_splits = input_159_split_num_splits_0, x = input_157_cast_fp16)[name = tensor("input_159_split_cast_fp16")]; + tensor input_159_split_1_sigmoid_cast_fp16 = sigmoid(x = input_159_split_cast_fp16_1)[name = tensor("input_159_split_1_sigmoid_cast_fp16")]; + tensor input_159_cast_fp16 = mul(x = input_159_split_cast_fp16_0, y = input_159_split_1_sigmoid_cast_fp16)[name = tensor("input_159_cast_fp16")]; + tensor input_161_pad_type_0 = const()[name = tensor("input_161_pad_type_0"), val = tensor("custom")]; + tensor input_161_pad_0 = const()[name = tensor("input_161_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_161_groups_0 = const()[name = tensor("input_161_groups_0"), val = tensor(1024)]; + tensor input_161_strides_0 = const()[name = tensor("input_161_strides_0"), val = tensor([1, 1])]; + tensor input_161_dilations_0 = const()[name = tensor("input_161_dilations_0"), val = tensor([1, 1])]; + tensor const_278_to_fp16 = const()[name = tensor("const_278_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134534336)))]; + tensor const_279_to_fp16 = const()[name = tensor("const_279_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134552832)))]; + tensor input_163_cast_fp16 = conv(bias = const_279_to_fp16, dilations = input_161_dilations_0, groups = input_161_groups_0, pad = input_161_pad_0, pad_type = input_161_pad_type_0, strides = input_161_strides_0, weight = const_278_to_fp16, x = input_159_cast_fp16)[name = tensor("input_163_cast_fp16")]; + tensor input_165_cast_fp16 = silu(x = input_163_cast_fp16)[name = tensor("input_165_cast_fp16")]; + tensor var_2340_pad_type_0 = const()[name = tensor("op_2340_pad_type_0"), val = tensor("valid")]; + tensor var_2340_strides_0 = const()[name = tensor("op_2340_strides_0"), val = tensor([1, 1])]; + tensor var_2340_pad_0 = const()[name = tensor("op_2340_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2340_dilations_0 = const()[name = tensor("op_2340_dilations_0"), val = tensor([1, 1])]; + tensor var_2340_groups_0 = const()[name = tensor("op_2340_groups_0"), val = tensor(1)]; + tensor layers_5_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134554944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135341440))), name = tensor("layers_5_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2340_cast_fp16 = conv(dilations = var_2340_dilations_0, groups = var_2340_groups_0, pad = var_2340_pad_0, pad_type = var_2340_pad_type_0, strides = var_2340_strides_0, weight = layers_5_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_165_cast_fp16)[name = tensor("op_2340_cast_fp16")]; + tensor var_2346_pad_type_0 = const()[name = tensor("op_2346_pad_type_0"), val = tensor("valid")]; + tensor var_2346_strides_0 = const()[name = tensor("op_2346_strides_0"), val = tensor([1, 1])]; + tensor var_2346_pad_0 = const()[name = tensor("op_2346_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2346_dilations_0 = const()[name = tensor("op_2346_dilations_0"), val = tensor([1, 1])]; + tensor var_2346_groups_0 = const()[name = tensor("op_2346_groups_0"), val = tensor(1)]; + tensor layers_5_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135373312))), name = tensor("layers_5_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135341632))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2346_cast_fp16 = conv(dilations = var_2346_dilations_0, groups = var_2346_groups_0, pad = var_2346_pad_0, pad_type = var_2346_pad_type_0, strides = var_2346_strides_0, weight = layers_5_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_165_cast_fp16)[name = tensor("op_2346_cast_fp16")]; + tensor x_35_cast_fp16 = add(x = var_2340_cast_fp16, y = var_2346_cast_fp16)[name = tensor("x_35_cast_fp16")]; + tensor inputs_57_cast_fp16 = add(x = inputs_55_cast_fp16, y = x_35_cast_fp16)[name = tensor("inputs_57_cast_fp16")]; + tensor out_57_axes_0 = const()[name = tensor("out_57_axes_0"), val = tensor([1])]; + tensor var_2357_to_fp16 = const()[name = tensor("op_2357_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_57_cast_fp16 = layer_norm(axes = out_57_axes_0, epsilon = var_2357_to_fp16, x = inputs_57_cast_fp16)[name = tensor("out_57_cast_fp16")]; + tensor input_167_gamma_0_to_fp16 = const()[name = tensor("input_167_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135504448)))]; + tensor input_167_beta_0_to_fp16 = const()[name = tensor("input_167_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135506560)))]; + tensor input_167_epsilon_0_to_fp16 = const()[name = tensor("input_167_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_167_cast_fp16 = batch_norm(beta = input_167_beta_0_to_fp16, epsilon = input_167_epsilon_0_to_fp16, gamma = input_167_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_57_cast_fp16)[name = tensor("input_167_cast_fp16")]; + tensor var_2377_pad_type_0 = const()[name = tensor("op_2377_pad_type_0"), val = tensor("valid")]; + tensor var_2377_strides_0 = const()[name = tensor("op_2377_strides_0"), val = tensor([1, 1])]; + tensor var_2377_pad_0 = const()[name = tensor("op_2377_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2377_dilations_0 = const()[name = tensor("op_2377_dilations_0"), val = tensor([1, 1])]; + tensor var_2377_groups_0 = const()[name = tensor("op_2377_groups_0"), val = tensor(1)]; + tensor layers_5_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135508672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138654464))), name = tensor("layers_5_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_2377_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_2377_dilations_0, groups = var_2377_groups_0, pad = var_2377_pad_0, pad_type = var_2377_pad_type_0, strides = var_2377_strides_0, weight = layers_5_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_167_cast_fp16)[name = tensor("op_2377_cast_fp16")]; + tensor var_2383_pad_type_0 = const()[name = tensor("op_2383_pad_type_0"), val = tensor("valid")]; + tensor var_2383_strides_0 = const()[name = tensor("op_2383_strides_0"), val = tensor([1, 1])]; + tensor var_2383_pad_0 = const()[name = tensor("op_2383_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2383_dilations_0 = const()[name = tensor("op_2383_dilations_0"), val = tensor([1, 1])]; + tensor var_2383_groups_0 = const()[name = tensor("op_2383_groups_0"), val = tensor(1)]; + tensor layers_5_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138807040))), name = tensor("layers_5_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138654656))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_2383_cast_fp16 = conv(dilations = var_2383_dilations_0, groups = var_2383_groups_0, pad = var_2383_pad_0, pad_type = var_2383_pad_type_0, strides = var_2383_strides_0, weight = layers_5_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_167_cast_fp16)[name = tensor("op_2383_cast_fp16")]; + tensor input_169_cast_fp16 = add(x = var_2377_cast_fp16, y = var_2383_cast_fp16)[name = tensor("input_169_cast_fp16")]; + tensor input_171_cast_fp16 = silu(x = input_169_cast_fp16)[name = tensor("input_171_cast_fp16")]; + tensor var_2394_pad_type_0 = const()[name = tensor("op_2394_pad_type_0"), val = tensor("valid")]; + tensor var_2394_strides_0 = const()[name = tensor("op_2394_strides_0"), val = tensor([1, 1])]; + tensor var_2394_pad_0 = const()[name = tensor("op_2394_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2394_dilations_0 = const()[name = tensor("op_2394_dilations_0"), val = tensor([1, 1])]; + tensor var_2394_groups_0 = const()[name = tensor("op_2394_groups_0"), val = tensor(1)]; + tensor layers_5_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(139331392))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142477184))), name = tensor("layers_5_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_2394_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2394_dilations_0, groups = var_2394_groups_0, pad = var_2394_pad_0, pad_type = var_2394_pad_type_0, strides = var_2394_strides_0, weight = layers_5_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_171_cast_fp16)[name = tensor("op_2394_cast_fp16")]; + tensor var_2400_pad_type_0 = const()[name = tensor("op_2400_pad_type_0"), val = tensor("valid")]; + tensor var_2400_strides_0 = const()[name = tensor("op_2400_strides_0"), val = tensor([1, 1])]; + tensor var_2400_pad_0 = const()[name = tensor("op_2400_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2400_dilations_0 = const()[name = tensor("op_2400_dilations_0"), val = tensor([1, 1])]; + tensor var_2400_groups_0 = const()[name = tensor("op_2400_groups_0"), val = tensor(1)]; + tensor layers_5_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142642048))), name = tensor("layers_5_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142477376))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_2400_cast_fp16 = conv(dilations = var_2400_dilations_0, groups = var_2400_groups_0, pad = var_2400_pad_0, pad_type = var_2400_pad_type_0, strides = var_2400_strides_0, weight = layers_5_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_171_cast_fp16)[name = tensor("op_2400_cast_fp16")]; + tensor x_37_cast_fp16 = add(x = var_2394_cast_fp16, y = var_2400_cast_fp16)[name = tensor("x_37_cast_fp16")]; + tensor var_2402_to_fp16 = const()[name = tensor("op_2402_to_fp16"), val = tensor(0x1p-1)]; + tensor var_2403_cast_fp16 = mul(x = x_37_cast_fp16, y = var_2402_to_fp16)[name = tensor("op_2403_cast_fp16")]; + tensor inputs_59_cast_fp16 = add(x = inputs_57_cast_fp16, y = var_2403_cast_fp16)[name = tensor("inputs_59_cast_fp16")]; + tensor out_59_axes_0 = const()[name = tensor("out_59_axes_0"), val = tensor([1])]; + tensor var_2413_to_fp16 = const()[name = tensor("op_2413_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_59_cast_fp16 = layer_norm(axes = out_59_axes_0, epsilon = var_2413_to_fp16, x = inputs_59_cast_fp16)[name = tensor("out_59_cast_fp16")]; + tensor inputs_61_gamma_0_to_fp16 = const()[name = tensor("inputs_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143166400)))]; + tensor inputs_61_beta_0_to_fp16 = const()[name = tensor("inputs_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143168512)))]; + tensor inputs_61_epsilon_0_to_fp16 = const()[name = tensor("inputs_61_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_61_cast_fp16 = batch_norm(beta = inputs_61_beta_0_to_fp16, epsilon = inputs_61_epsilon_0_to_fp16, gamma = inputs_61_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_59_cast_fp16)[name = tensor("inputs_61_cast_fp16")]; + tensor var_2427 = const()[name = tensor("op_2427"), val = tensor(3)]; + tensor out_61_axes_0 = const()[name = tensor("out_61_axes_0"), val = tensor([1])]; + tensor var_2458_to_fp16 = const()[name = tensor("op_2458_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_61_cast_fp16 = layer_norm(axes = out_61_axes_0, epsilon = var_2458_to_fp16, x = inputs_61_cast_fp16)[name = tensor("out_61_cast_fp16")]; + tensor input_173_gamma_0_to_fp16 = const()[name = tensor("input_173_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143170624)))]; + tensor input_173_beta_0_to_fp16 = const()[name = tensor("input_173_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143172736)))]; + tensor input_173_epsilon_0_to_fp16 = const()[name = tensor("input_173_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_173_cast_fp16 = batch_norm(beta = input_173_beta_0_to_fp16, epsilon = input_173_epsilon_0_to_fp16, gamma = input_173_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_61_cast_fp16)[name = tensor("input_173_cast_fp16")]; + tensor var_2478_pad_type_0 = const()[name = tensor("op_2478_pad_type_0"), val = tensor("valid")]; + tensor var_2478_strides_0 = const()[name = tensor("op_2478_strides_0"), val = tensor([1, 1])]; + tensor var_2478_pad_0 = const()[name = tensor("op_2478_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2478_dilations_0 = const()[name = tensor("op_2478_dilations_0"), val = tensor([1, 1])]; + tensor var_2478_groups_0 = const()[name = tensor("op_2478_groups_0"), val = tensor(1)]; + tensor layers_6_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143174848))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(146320640))), name = tensor("layers_6_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_2478_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_2478_dilations_0, groups = var_2478_groups_0, pad = var_2478_pad_0, pad_type = var_2478_pad_type_0, strides = var_2478_strides_0, weight = layers_6_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_173_cast_fp16)[name = tensor("op_2478_cast_fp16")]; + tensor var_2484_pad_type_0 = const()[name = tensor("op_2484_pad_type_0"), val = tensor("valid")]; + tensor var_2484_strides_0 = const()[name = tensor("op_2484_strides_0"), val = tensor([1, 1])]; + tensor var_2484_pad_0 = const()[name = tensor("op_2484_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2484_dilations_0 = const()[name = tensor("op_2484_dilations_0"), val = tensor([1, 1])]; + tensor var_2484_groups_0 = const()[name = tensor("op_2484_groups_0"), val = tensor(1)]; + tensor layers_6_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(146465152))), name = tensor("layers_6_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(146320832))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_2484_cast_fp16 = conv(dilations = var_2484_dilations_0, groups = var_2484_groups_0, pad = var_2484_pad_0, pad_type = var_2484_pad_type_0, strides = var_2484_strides_0, weight = layers_6_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_173_cast_fp16)[name = tensor("op_2484_cast_fp16")]; + tensor input_175_cast_fp16 = add(x = var_2478_cast_fp16, y = var_2484_cast_fp16)[name = tensor("input_175_cast_fp16")]; + tensor input_177_cast_fp16 = silu(x = input_175_cast_fp16)[name = tensor("input_177_cast_fp16")]; + tensor var_2495_pad_type_0 = const()[name = tensor("op_2495_pad_type_0"), val = tensor("valid")]; + tensor var_2495_strides_0 = const()[name = tensor("op_2495_strides_0"), val = tensor([1, 1])]; + tensor var_2495_pad_0 = const()[name = tensor("op_2495_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2495_dilations_0 = const()[name = tensor("op_2495_dilations_0"), val = tensor([1, 1])]; + tensor var_2495_groups_0 = const()[name = tensor("op_2495_groups_0"), val = tensor(1)]; + tensor layers_6_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(146989504))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150135296))), name = tensor("layers_6_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_2495_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2495_dilations_0, groups = var_2495_groups_0, pad = var_2495_pad_0, pad_type = var_2495_pad_type_0, strides = var_2495_strides_0, weight = layers_6_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_177_cast_fp16)[name = tensor("op_2495_cast_fp16")]; + tensor var_2501_pad_type_0 = const()[name = tensor("op_2501_pad_type_0"), val = tensor("valid")]; + tensor var_2501_strides_0 = const()[name = tensor("op_2501_strides_0"), val = tensor([1, 1])]; + tensor var_2501_pad_0 = const()[name = tensor("op_2501_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2501_dilations_0 = const()[name = tensor("op_2501_dilations_0"), val = tensor([1, 1])]; + tensor var_2501_groups_0 = const()[name = tensor("op_2501_groups_0"), val = tensor(1)]; + tensor layers_6_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150314688))), name = tensor("layers_6_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150135488))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_2501_cast_fp16 = conv(dilations = var_2501_dilations_0, groups = var_2501_groups_0, pad = var_2501_pad_0, pad_type = var_2501_pad_type_0, strides = var_2501_strides_0, weight = layers_6_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_177_cast_fp16)[name = tensor("op_2501_cast_fp16")]; + tensor x_39_cast_fp16 = add(x = var_2495_cast_fp16, y = var_2501_cast_fp16)[name = tensor("x_39_cast_fp16")]; + tensor var_2503_to_fp16 = const()[name = tensor("op_2503_to_fp16"), val = tensor(0x1p-1)]; + tensor var_2504_cast_fp16 = mul(x = x_39_cast_fp16, y = var_2503_to_fp16)[name = tensor("op_2504_cast_fp16")]; + tensor inputs_63_cast_fp16 = add(x = inputs_61_cast_fp16, y = var_2504_cast_fp16)[name = tensor("inputs_63_cast_fp16")]; + tensor out_63_axes_0 = const()[name = tensor("out_63_axes_0"), val = tensor([1])]; + tensor var_2514_to_fp16 = const()[name = tensor("op_2514_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_63_cast_fp16 = layer_norm(axes = out_63_axes_0, epsilon = var_2514_to_fp16, x = inputs_63_cast_fp16)[name = tensor("out_63_cast_fp16")]; + tensor obj_27_gamma_0_to_fp16 = const()[name = tensor("obj_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150839040)))]; + tensor obj_27_beta_0_to_fp16 = const()[name = tensor("obj_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150841152)))]; + tensor obj_27_epsilon_0_to_fp16 = const()[name = tensor("obj_27_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_27_cast_fp16 = batch_norm(beta = obj_27_beta_0_to_fp16, epsilon = obj_27_epsilon_0_to_fp16, gamma = obj_27_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_63_cast_fp16)[name = tensor("obj_27_cast_fp16")]; + tensor var_2539_pad_type_0 = const()[name = tensor("op_2539_pad_type_0"), val = tensor("valid")]; + tensor var_2539_strides_0 = const()[name = tensor("op_2539_strides_0"), val = tensor([1, 1])]; + tensor var_2539_pad_0 = const()[name = tensor("op_2539_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2539_dilations_0 = const()[name = tensor("op_2539_dilations_0"), val = tensor([1, 1])]; + tensor var_2539_groups_0 = const()[name = tensor("op_2539_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150843264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151629760))), name = tensor("layers_6_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2539_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2539_dilations_0, groups = var_2539_groups_0, pad = var_2539_pad_0, pad_type = var_2539_pad_type_0, strides = var_2539_strides_0, weight = layers_6_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_27_cast_fp16)[name = tensor("op_2539_cast_fp16")]; + tensor var_2545_pad_type_0 = const()[name = tensor("op_2545_pad_type_0"), val = tensor("valid")]; + tensor var_2545_strides_0 = const()[name = tensor("op_2545_strides_0"), val = tensor([1, 1])]; + tensor var_2545_pad_0 = const()[name = tensor("op_2545_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2545_dilations_0 = const()[name = tensor("op_2545_dilations_0"), val = tensor([1, 1])]; + tensor var_2545_groups_0 = const()[name = tensor("op_2545_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151664256))), name = tensor("layers_6_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151629952))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2545_cast_fp16 = conv(dilations = var_2545_dilations_0, groups = var_2545_groups_0, pad = var_2545_pad_0, pad_type = var_2545_pad_type_0, strides = var_2545_strides_0, weight = layers_6_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_27_cast_fp16)[name = tensor("op_2545_cast_fp16")]; + tensor query_25_cast_fp16 = add(x = var_2539_cast_fp16, y = var_2545_cast_fp16)[name = tensor("query_25_cast_fp16")]; + tensor var_2554_pad_type_0 = const()[name = tensor("op_2554_pad_type_0"), val = tensor("valid")]; + tensor var_2554_strides_0 = const()[name = tensor("op_2554_strides_0"), val = tensor([1, 1])]; + tensor var_2554_pad_0 = const()[name = tensor("op_2554_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2554_dilations_0 = const()[name = tensor("op_2554_dilations_0"), val = tensor([1, 1])]; + tensor var_2554_groups_0 = const()[name = tensor("op_2554_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151795392))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152581888))), name = tensor("layers_6_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2554_cast_fp16 = conv(dilations = var_2554_dilations_0, groups = var_2554_groups_0, pad = var_2554_pad_0, pad_type = var_2554_pad_type_0, strides = var_2554_strides_0, weight = layers_6_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_27_cast_fp16)[name = tensor("op_2554_cast_fp16")]; + tensor var_2560_pad_type_0 = const()[name = tensor("op_2560_pad_type_0"), val = tensor("valid")]; + tensor var_2560_strides_0 = const()[name = tensor("op_2560_strides_0"), val = tensor([1, 1])]; + tensor var_2560_pad_0 = const()[name = tensor("op_2560_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2560_dilations_0 = const()[name = tensor("op_2560_dilations_0"), val = tensor([1, 1])]; + tensor var_2560_groups_0 = const()[name = tensor("op_2560_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152625920))), name = tensor("layers_6_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152582080))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2560_cast_fp16 = conv(dilations = var_2560_dilations_0, groups = var_2560_groups_0, pad = var_2560_pad_0, pad_type = var_2560_pad_type_0, strides = var_2560_strides_0, weight = layers_6_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_27_cast_fp16)[name = tensor("op_2560_cast_fp16")]; + tensor key_13_cast_fp16 = add(x = var_2554_cast_fp16, y = var_2560_cast_fp16)[name = tensor("key_13_cast_fp16")]; + tensor var_2570_pad_type_0 = const()[name = tensor("op_2570_pad_type_0"), val = tensor("valid")]; + tensor var_2570_strides_0 = const()[name = tensor("op_2570_strides_0"), val = tensor([1, 1])]; + tensor var_2570_pad_0 = const()[name = tensor("op_2570_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2570_dilations_0 = const()[name = tensor("op_2570_dilations_0"), val = tensor([1, 1])]; + tensor var_2570_groups_0 = const()[name = tensor("op_2570_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152757056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(153543552))), name = tensor("layers_6_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2570_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2570_dilations_0, groups = var_2570_groups_0, pad = var_2570_pad_0, pad_type = var_2570_pad_type_0, strides = var_2570_strides_0, weight = layers_6_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_27_cast_fp16)[name = tensor("op_2570_cast_fp16")]; + tensor var_2576_pad_type_0 = const()[name = tensor("op_2576_pad_type_0"), val = tensor("valid")]; + tensor var_2576_strides_0 = const()[name = tensor("op_2576_strides_0"), val = tensor([1, 1])]; + tensor var_2576_pad_0 = const()[name = tensor("op_2576_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2576_dilations_0 = const()[name = tensor("op_2576_dilations_0"), val = tensor([1, 1])]; + tensor var_2576_groups_0 = const()[name = tensor("op_2576_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(153577024))), name = tensor("layers_6_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(153543744))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2576_cast_fp16 = conv(dilations = var_2576_dilations_0, groups = var_2576_groups_0, pad = var_2576_pad_0, pad_type = var_2576_pad_type_0, strides = var_2576_strides_0, weight = layers_6_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_27_cast_fp16)[name = tensor("op_2576_cast_fp16")]; + tensor value_13_cast_fp16 = add(x = var_2570_cast_fp16, y = var_2576_cast_fp16)[name = tensor("value_13_cast_fp16")]; + tensor var_2579_to_fp16 = const()[name = tensor("op_2579_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(153708160)))]; + tensor query_27_cast_fp16 = add(x = query_25_cast_fp16, y = var_2579_to_fp16)[name = tensor("query_27_cast_fp16")]; + tensor var_2582_to_fp16 = const()[name = tensor("op_2582_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(153710272)))]; + tensor q_with_bias_v_13_cast_fp16 = add(x = query_25_cast_fp16, y = var_2582_to_fp16)[name = tensor("q_with_bias_v_13_cast_fp16")]; + tensor var_2592_pad_type_0 = const()[name = tensor("op_2592_pad_type_0"), val = tensor("valid")]; + tensor var_2592_strides_0 = const()[name = tensor("op_2592_strides_0"), val = tensor([1, 1])]; + tensor var_2592_pad_0 = const()[name = tensor("op_2592_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2592_dilations_0 = const()[name = tensor("op_2592_dilations_0"), val = tensor([1, 1])]; + tensor var_2592_groups_0 = const()[name = tensor("op_2592_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(153712384))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(154498880))), name = tensor("layers_6_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2592_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2592_dilations_0, groups = var_2592_groups_0, pad = var_2592_pad_0, pad_type = var_2592_pad_type_0, strides = var_2592_strides_0, weight = layers_6_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_2592_cast_fp16")]; + tensor var_2598_pad_type_0 = const()[name = tensor("op_2598_pad_type_0"), val = tensor("valid")]; + tensor var_2598_strides_0 = const()[name = tensor("op_2598_strides_0"), val = tensor([1, 1])]; + tensor var_2598_pad_0 = const()[name = tensor("op_2598_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2598_dilations_0 = const()[name = tensor("op_2598_dilations_0"), val = tensor([1, 1])]; + tensor var_2598_groups_0 = const()[name = tensor("op_2598_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(154578496))), name = tensor("layers_6_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(154499072))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2598_cast_fp16 = conv(dilations = var_2598_dilations_0, groups = var_2598_groups_0, pad = var_2598_pad_0, pad_type = var_2598_pad_type_0, strides = var_2598_strides_0, weight = layers_6_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_2598_cast_fp16")]; + tensor p_13_cast_fp16 = add(x = var_2592_cast_fp16, y = var_2598_cast_fp16)[name = tensor("p_13_cast_fp16")]; + tensor var_2602 = const()[name = tensor("op_2602"), val = tensor([1, 8, 128, 188])]; + tensor var_2603_cast_fp16 = reshape(shape = var_2602, x = q_with_bias_v_13_cast_fp16)[name = tensor("op_2603_cast_fp16")]; + tensor var_2604 = const()[name = tensor("op_2604"), val = tensor([1, 8, 128, -1])]; + tensor var_2605_cast_fp16 = reshape(shape = var_2604, x = p_13_cast_fp16)[name = tensor("op_2605_cast_fp16")]; + tensor matrix_bd_49_transpose_x_0 = const()[name = tensor("matrix_bd_49_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_49_transpose_y_0 = const()[name = tensor("matrix_bd_49_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_49_cast_fp16 = matmul(transpose_x = matrix_bd_49_transpose_x_0, transpose_y = matrix_bd_49_transpose_y_0, x = var_2603_cast_fp16, y = var_2605_cast_fp16)[name = tensor("matrix_bd_49_cast_fp16")]; + tensor matrix_bd_51_pad_0 = const()[name = tensor("matrix_bd_51_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_51_mode_0 = const()[name = tensor("matrix_bd_51_mode_0"), val = tensor("constant")]; + tensor const_76_to_fp16 = const()[name = tensor("const_76_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_51_cast_fp16 = pad(constant_val = const_76_to_fp16, mode = matrix_bd_51_mode_0, pad = matrix_bd_51_pad_0, x = matrix_bd_49_cast_fp16)[name = tensor("matrix_bd_51_cast_fp16")]; + tensor var_2614 = const()[name = tensor("op_2614"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_53_cast_fp16 = reshape(shape = var_2614, x = matrix_bd_51_cast_fp16)[name = tensor("matrix_bd_53_cast_fp16")]; + tensor var_2618_begin_0 = const()[name = tensor("op_2618_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_2618_end_0 = const()[name = tensor("op_2618_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_2618_end_mask_0 = const()[name = tensor("op_2618_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2618_cast_fp16 = slice_by_index(begin = var_2618_begin_0, end = var_2618_end_0, end_mask = var_2618_end_mask_0, x = matrix_bd_53_cast_fp16)[name = tensor("op_2618_cast_fp16")]; + tensor var_2619 = const()[name = tensor("op_2619"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_55_cast_fp16 = reshape(shape = var_2619, x = var_2618_cast_fp16)[name = tensor("matrix_bd_55_cast_fp16")]; + tensor var_2624_begin_0 = const()[name = tensor("op_2624_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2624_end_0 = const()[name = tensor("op_2624_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_2624_end_mask_0 = const()[name = tensor("op_2624_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2624_cast_fp16 = slice_by_index(begin = var_2624_begin_0, end = var_2624_end_0, end_mask = var_2624_end_mask_0, x = matrix_bd_55_cast_fp16)[name = tensor("op_2624_cast_fp16")]; + tensor var_2625_to_fp16 = const()[name = tensor("op_2625_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_13_cast_fp16 = mul(x = var_2624_cast_fp16, y = var_2625_to_fp16)[name = tensor("qk_mask_13_cast_fp16")]; + tensor var_2629 = const()[name = tensor("op_2629"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_13_cast_fp16 = reshape(shape = var_2629, x = query_27_cast_fp16)[name = tensor("mh_q_13_cast_fp16")]; + tensor var_2631_to_fp16 = const()[name = tensor("op_2631_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_2632_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_2631_to_fp16)[name = tensor("op_2632_cast_fp16")]; + tensor var_2635 = const()[name = tensor("op_2635"), val = tensor([1, 8, 128, 188])]; + tensor var_2636_cast_fp16 = reshape(shape = var_2635, x = key_13_cast_fp16)[name = tensor("op_2636_cast_fp16")]; + tensor mh_w_25_transpose_x_0 = const()[name = tensor("mh_w_25_transpose_x_0"), val = tensor(true)]; + tensor mh_w_25_transpose_y_0 = const()[name = tensor("mh_w_25_transpose_y_0"), val = tensor(false)]; + tensor mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_2632_cast_fp16, y = var_2636_cast_fp16)[name = tensor("mh_w_25_cast_fp16")]; + tensor mh_w_27_cast_fp16 = add(x = mh_w_25_cast_fp16, y = qk_mask_13_cast_fp16)[name = tensor("mh_w_27_cast_fp16")]; + tensor var_2640_cast_fp16 = softmax(axis = var_2427, x = mh_w_27_cast_fp16)[name = tensor("op_2640_cast_fp16")]; + tensor var_2641 = const()[name = tensor("op_2641"), val = tensor([1, 8, 128, 188])]; + tensor var_2642_cast_fp16 = reshape(shape = var_2641, x = value_13_cast_fp16)[name = tensor("op_2642_cast_fp16")]; + tensor attn_13_transpose_x_0 = const()[name = tensor("attn_13_transpose_x_0"), val = tensor(false)]; + tensor attn_13_transpose_y_0 = const()[name = tensor("attn_13_transpose_y_0"), val = tensor(true)]; + tensor attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_2642_cast_fp16, y = var_2640_cast_fp16)[name = tensor("attn_13_cast_fp16")]; + tensor var_2645 = const()[name = tensor("op_2645"), val = tensor([1, 1024, 1, 188])]; + tensor input_179_cast_fp16 = reshape(shape = var_2645, x = attn_13_cast_fp16)[name = tensor("input_179_cast_fp16")]; + tensor var_2655_pad_type_0 = const()[name = tensor("op_2655_pad_type_0"), val = tensor("valid")]; + tensor var_2655_strides_0 = const()[name = tensor("op_2655_strides_0"), val = tensor([1, 1])]; + tensor var_2655_pad_0 = const()[name = tensor("op_2655_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2655_dilations_0 = const()[name = tensor("op_2655_dilations_0"), val = tensor([1, 1])]; + tensor var_2655_groups_0 = const()[name = tensor("op_2655_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(154709632))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(155496128))), name = tensor("layers_6_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2655_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2655_dilations_0, groups = var_2655_groups_0, pad = var_2655_pad_0, pad_type = var_2655_pad_type_0, strides = var_2655_strides_0, weight = layers_6_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_179_cast_fp16)[name = tensor("op_2655_cast_fp16")]; + tensor var_2661_pad_type_0 = const()[name = tensor("op_2661_pad_type_0"), val = tensor("valid")]; + tensor var_2661_strides_0 = const()[name = tensor("op_2661_strides_0"), val = tensor([1, 1])]; + tensor var_2661_pad_0 = const()[name = tensor("op_2661_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2661_dilations_0 = const()[name = tensor("op_2661_dilations_0"), val = tensor([1, 1])]; + tensor var_2661_groups_0 = const()[name = tensor("op_2661_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(155526912))), name = tensor("layers_6_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(155496320))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2661_cast_fp16 = conv(dilations = var_2661_dilations_0, groups = var_2661_groups_0, pad = var_2661_pad_0, pad_type = var_2661_pad_type_0, strides = var_2661_strides_0, weight = layers_6_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_179_cast_fp16)[name = tensor("op_2661_cast_fp16")]; + tensor obj_29_cast_fp16 = add(x = var_2655_cast_fp16, y = var_2661_cast_fp16)[name = tensor("obj_29_cast_fp16")]; + tensor inputs_65_cast_fp16 = add(x = inputs_63_cast_fp16, y = obj_29_cast_fp16)[name = tensor("inputs_65_cast_fp16")]; + tensor out_65_axes_0 = const()[name = tensor("out_65_axes_0"), val = tensor([1])]; + tensor var_2672_to_fp16 = const()[name = tensor("op_2672_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_65_cast_fp16 = layer_norm(axes = out_65_axes_0, epsilon = var_2672_to_fp16, x = inputs_65_cast_fp16)[name = tensor("out_65_cast_fp16")]; + tensor input_181_gamma_0_to_fp16 = const()[name = tensor("input_181_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(155658048)))]; + tensor input_181_beta_0_to_fp16 = const()[name = tensor("input_181_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(155660160)))]; + tensor input_181_epsilon_0_to_fp16 = const()[name = tensor("input_181_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_181_cast_fp16 = batch_norm(beta = input_181_beta_0_to_fp16, epsilon = input_181_epsilon_0_to_fp16, gamma = input_181_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_65_cast_fp16)[name = tensor("input_181_cast_fp16")]; + tensor var_2693_pad_type_0 = const()[name = tensor("op_2693_pad_type_0"), val = tensor("valid")]; + tensor var_2693_strides_0 = const()[name = tensor("op_2693_strides_0"), val = tensor([1, 1])]; + tensor var_2693_pad_0 = const()[name = tensor("op_2693_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2693_dilations_0 = const()[name = tensor("op_2693_dilations_0"), val = tensor([1, 1])]; + tensor var_2693_groups_0 = const()[name = tensor("op_2693_groups_0"), val = tensor(1)]; + tensor layers_6_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(155662272))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(157235200))), name = tensor("layers_6_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_2693_cast_fp16 = conv(dilations = var_2693_dilations_0, groups = var_2693_groups_0, pad = var_2693_pad_0, pad_type = var_2693_pad_type_0, strides = var_2693_strides_0, weight = layers_6_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_181_cast_fp16)[name = tensor("op_2693_cast_fp16")]; + tensor var_2699_pad_type_0 = const()[name = tensor("op_2699_pad_type_0"), val = tensor("valid")]; + tensor var_2699_strides_0 = const()[name = tensor("op_2699_strides_0"), val = tensor([1, 1])]; + tensor var_2699_pad_0 = const()[name = tensor("op_2699_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2699_dilations_0 = const()[name = tensor("op_2699_dilations_0"), val = tensor([1, 1])]; + tensor var_2699_groups_0 = const()[name = tensor("op_2699_groups_0"), val = tensor(1)]; + tensor layers_6_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(157299584))), name = tensor("layers_6_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(157235392))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_2699_cast_fp16 = conv(dilations = var_2699_dilations_0, groups = var_2699_groups_0, pad = var_2699_pad_0, pad_type = var_2699_pad_type_0, strides = var_2699_strides_0, weight = layers_6_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_181_cast_fp16)[name = tensor("op_2699_cast_fp16")]; + tensor input_183_cast_fp16 = add(x = var_2693_cast_fp16, y = var_2699_cast_fp16)[name = tensor("input_183_cast_fp16")]; + tensor input_185_split_num_splits_0 = const()[name = tensor("input_185_split_num_splits_0"), val = tensor(2)]; + tensor input_185_split_axis_0 = const()[name = tensor("input_185_split_axis_0"), val = tensor(1)]; + tensor input_185_split_cast_fp16_0, tensor input_185_split_cast_fp16_1 = split(axis = input_185_split_axis_0, num_splits = input_185_split_num_splits_0, x = input_183_cast_fp16)[name = tensor("input_185_split_cast_fp16")]; + tensor input_185_split_1_sigmoid_cast_fp16 = sigmoid(x = input_185_split_cast_fp16_1)[name = tensor("input_185_split_1_sigmoid_cast_fp16")]; + tensor input_185_cast_fp16 = mul(x = input_185_split_cast_fp16_0, y = input_185_split_1_sigmoid_cast_fp16)[name = tensor("input_185_cast_fp16")]; + tensor input_187_pad_type_0 = const()[name = tensor("input_187_pad_type_0"), val = tensor("custom")]; + tensor input_187_pad_0 = const()[name = tensor("input_187_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_187_groups_0 = const()[name = tensor("input_187_groups_0"), val = tensor(1024)]; + tensor input_187_strides_0 = const()[name = tensor("input_187_strides_0"), val = tensor([1, 1])]; + tensor input_187_dilations_0 = const()[name = tensor("input_187_dilations_0"), val = tensor([1, 1])]; + tensor const_280_to_fp16 = const()[name = tensor("const_280_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(157561792)))]; + tensor const_281_to_fp16 = const()[name = tensor("const_281_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(157580288)))]; + tensor input_189_cast_fp16 = conv(bias = const_281_to_fp16, dilations = input_187_dilations_0, groups = input_187_groups_0, pad = input_187_pad_0, pad_type = input_187_pad_type_0, strides = input_187_strides_0, weight = const_280_to_fp16, x = input_185_cast_fp16)[name = tensor("input_189_cast_fp16")]; + tensor input_191_cast_fp16 = silu(x = input_189_cast_fp16)[name = tensor("input_191_cast_fp16")]; + tensor var_2721_pad_type_0 = const()[name = tensor("op_2721_pad_type_0"), val = tensor("valid")]; + tensor var_2721_strides_0 = const()[name = tensor("op_2721_strides_0"), val = tensor([1, 1])]; + tensor var_2721_pad_0 = const()[name = tensor("op_2721_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2721_dilations_0 = const()[name = tensor("op_2721_dilations_0"), val = tensor([1, 1])]; + tensor var_2721_groups_0 = const()[name = tensor("op_2721_groups_0"), val = tensor(1)]; + tensor layers_6_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(157582400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158368896))), name = tensor("layers_6_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2721_cast_fp16 = conv(dilations = var_2721_dilations_0, groups = var_2721_groups_0, pad = var_2721_pad_0, pad_type = var_2721_pad_type_0, strides = var_2721_strides_0, weight = layers_6_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_191_cast_fp16)[name = tensor("op_2721_cast_fp16")]; + tensor var_2727_pad_type_0 = const()[name = tensor("op_2727_pad_type_0"), val = tensor("valid")]; + tensor var_2727_strides_0 = const()[name = tensor("op_2727_strides_0"), val = tensor([1, 1])]; + tensor var_2727_pad_0 = const()[name = tensor("op_2727_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2727_dilations_0 = const()[name = tensor("op_2727_dilations_0"), val = tensor([1, 1])]; + tensor var_2727_groups_0 = const()[name = tensor("op_2727_groups_0"), val = tensor(1)]; + tensor layers_6_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158400512))), name = tensor("layers_6_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158369088))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2727_cast_fp16 = conv(dilations = var_2727_dilations_0, groups = var_2727_groups_0, pad = var_2727_pad_0, pad_type = var_2727_pad_type_0, strides = var_2727_strides_0, weight = layers_6_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_191_cast_fp16)[name = tensor("op_2727_cast_fp16")]; + tensor x_41_cast_fp16 = add(x = var_2721_cast_fp16, y = var_2727_cast_fp16)[name = tensor("x_41_cast_fp16")]; + tensor inputs_67_cast_fp16 = add(x = inputs_65_cast_fp16, y = x_41_cast_fp16)[name = tensor("inputs_67_cast_fp16")]; + tensor out_67_axes_0 = const()[name = tensor("out_67_axes_0"), val = tensor([1])]; + tensor var_2738_to_fp16 = const()[name = tensor("op_2738_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_67_cast_fp16 = layer_norm(axes = out_67_axes_0, epsilon = var_2738_to_fp16, x = inputs_67_cast_fp16)[name = tensor("out_67_cast_fp16")]; + tensor input_193_gamma_0_to_fp16 = const()[name = tensor("input_193_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158531648)))]; + tensor input_193_beta_0_to_fp16 = const()[name = tensor("input_193_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158533760)))]; + tensor input_193_epsilon_0_to_fp16 = const()[name = tensor("input_193_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_193_cast_fp16 = batch_norm(beta = input_193_beta_0_to_fp16, epsilon = input_193_epsilon_0_to_fp16, gamma = input_193_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_67_cast_fp16)[name = tensor("input_193_cast_fp16")]; + tensor var_2758_pad_type_0 = const()[name = tensor("op_2758_pad_type_0"), val = tensor("valid")]; + tensor var_2758_strides_0 = const()[name = tensor("op_2758_strides_0"), val = tensor([1, 1])]; + tensor var_2758_pad_0 = const()[name = tensor("op_2758_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2758_dilations_0 = const()[name = tensor("op_2758_dilations_0"), val = tensor([1, 1])]; + tensor var_2758_groups_0 = const()[name = tensor("op_2758_groups_0"), val = tensor(1)]; + tensor layers_6_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158535872))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161681664))), name = tensor("layers_6_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_2758_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_2758_dilations_0, groups = var_2758_groups_0, pad = var_2758_pad_0, pad_type = var_2758_pad_type_0, strides = var_2758_strides_0, weight = layers_6_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_193_cast_fp16)[name = tensor("op_2758_cast_fp16")]; + tensor var_2764_pad_type_0 = const()[name = tensor("op_2764_pad_type_0"), val = tensor("valid")]; + tensor var_2764_strides_0 = const()[name = tensor("op_2764_strides_0"), val = tensor([1, 1])]; + tensor var_2764_pad_0 = const()[name = tensor("op_2764_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2764_dilations_0 = const()[name = tensor("op_2764_dilations_0"), val = tensor([1, 1])]; + tensor var_2764_groups_0 = const()[name = tensor("op_2764_groups_0"), val = tensor(1)]; + tensor layers_6_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161826432))), name = tensor("layers_6_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161681856))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_2764_cast_fp16 = conv(dilations = var_2764_dilations_0, groups = var_2764_groups_0, pad = var_2764_pad_0, pad_type = var_2764_pad_type_0, strides = var_2764_strides_0, weight = layers_6_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_193_cast_fp16)[name = tensor("op_2764_cast_fp16")]; + tensor input_195_cast_fp16 = add(x = var_2758_cast_fp16, y = var_2764_cast_fp16)[name = tensor("input_195_cast_fp16")]; + tensor input_197_cast_fp16 = silu(x = input_195_cast_fp16)[name = tensor("input_197_cast_fp16")]; + tensor var_2775_pad_type_0 = const()[name = tensor("op_2775_pad_type_0"), val = tensor("valid")]; + tensor var_2775_strides_0 = const()[name = tensor("op_2775_strides_0"), val = tensor([1, 1])]; + tensor var_2775_pad_0 = const()[name = tensor("op_2775_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2775_dilations_0 = const()[name = tensor("op_2775_dilations_0"), val = tensor([1, 1])]; + tensor var_2775_groups_0 = const()[name = tensor("op_2775_groups_0"), val = tensor(1)]; + tensor layers_6_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162350784))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165496576))), name = tensor("layers_6_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_2775_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2775_dilations_0, groups = var_2775_groups_0, pad = var_2775_pad_0, pad_type = var_2775_pad_type_0, strides = var_2775_strides_0, weight = layers_6_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_197_cast_fp16)[name = tensor("op_2775_cast_fp16")]; + tensor var_2781_pad_type_0 = const()[name = tensor("op_2781_pad_type_0"), val = tensor("valid")]; + tensor var_2781_strides_0 = const()[name = tensor("op_2781_strides_0"), val = tensor([1, 1])]; + tensor var_2781_pad_0 = const()[name = tensor("op_2781_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2781_dilations_0 = const()[name = tensor("op_2781_dilations_0"), val = tensor([1, 1])]; + tensor var_2781_groups_0 = const()[name = tensor("op_2781_groups_0"), val = tensor(1)]; + tensor layers_6_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165648256))), name = tensor("layers_6_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165496768))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_2781_cast_fp16 = conv(dilations = var_2781_dilations_0, groups = var_2781_groups_0, pad = var_2781_pad_0, pad_type = var_2781_pad_type_0, strides = var_2781_strides_0, weight = layers_6_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_197_cast_fp16)[name = tensor("op_2781_cast_fp16")]; + tensor x_43_cast_fp16 = add(x = var_2775_cast_fp16, y = var_2781_cast_fp16)[name = tensor("x_43_cast_fp16")]; + tensor var_2783_to_fp16 = const()[name = tensor("op_2783_to_fp16"), val = tensor(0x1p-1)]; + tensor var_2784_cast_fp16 = mul(x = x_43_cast_fp16, y = var_2783_to_fp16)[name = tensor("op_2784_cast_fp16")]; + tensor inputs_69_cast_fp16 = add(x = inputs_67_cast_fp16, y = var_2784_cast_fp16)[name = tensor("inputs_69_cast_fp16")]; + tensor out_69_axes_0 = const()[name = tensor("out_69_axes_0"), val = tensor([1])]; + tensor var_2794_to_fp16 = const()[name = tensor("op_2794_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_69_cast_fp16 = layer_norm(axes = out_69_axes_0, epsilon = var_2794_to_fp16, x = inputs_69_cast_fp16)[name = tensor("out_69_cast_fp16")]; + tensor inputs_71_gamma_0_to_fp16 = const()[name = tensor("inputs_71_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166172608)))]; + tensor inputs_71_beta_0_to_fp16 = const()[name = tensor("inputs_71_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166174720)))]; + tensor inputs_71_epsilon_0_to_fp16 = const()[name = tensor("inputs_71_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_71_cast_fp16 = batch_norm(beta = inputs_71_beta_0_to_fp16, epsilon = inputs_71_epsilon_0_to_fp16, gamma = inputs_71_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_69_cast_fp16)[name = tensor("inputs_71_cast_fp16")]; + tensor var_2808 = const()[name = tensor("op_2808"), val = tensor(3)]; + tensor out_71_axes_0 = const()[name = tensor("out_71_axes_0"), val = tensor([1])]; + tensor var_2839_to_fp16 = const()[name = tensor("op_2839_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_71_cast_fp16 = layer_norm(axes = out_71_axes_0, epsilon = var_2839_to_fp16, x = inputs_71_cast_fp16)[name = tensor("out_71_cast_fp16")]; + tensor input_199_gamma_0_to_fp16 = const()[name = tensor("input_199_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166176832)))]; + tensor input_199_beta_0_to_fp16 = const()[name = tensor("input_199_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166178944)))]; + tensor input_199_epsilon_0_to_fp16 = const()[name = tensor("input_199_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_199_cast_fp16 = batch_norm(beta = input_199_beta_0_to_fp16, epsilon = input_199_epsilon_0_to_fp16, gamma = input_199_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_71_cast_fp16)[name = tensor("input_199_cast_fp16")]; + tensor var_2859_pad_type_0 = const()[name = tensor("op_2859_pad_type_0"), val = tensor("valid")]; + tensor var_2859_strides_0 = const()[name = tensor("op_2859_strides_0"), val = tensor([1, 1])]; + tensor var_2859_pad_0 = const()[name = tensor("op_2859_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2859_dilations_0 = const()[name = tensor("op_2859_dilations_0"), val = tensor([1, 1])]; + tensor var_2859_groups_0 = const()[name = tensor("op_2859_groups_0"), val = tensor(1)]; + tensor layers_7_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166181056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169326848))), name = tensor("layers_7_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_2859_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_2859_dilations_0, groups = var_2859_groups_0, pad = var_2859_pad_0, pad_type = var_2859_pad_type_0, strides = var_2859_strides_0, weight = layers_7_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_199_cast_fp16)[name = tensor("op_2859_cast_fp16")]; + tensor var_2865_pad_type_0 = const()[name = tensor("op_2865_pad_type_0"), val = tensor("valid")]; + tensor var_2865_strides_0 = const()[name = tensor("op_2865_strides_0"), val = tensor([1, 1])]; + tensor var_2865_pad_0 = const()[name = tensor("op_2865_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2865_dilations_0 = const()[name = tensor("op_2865_dilations_0"), val = tensor([1, 1])]; + tensor var_2865_groups_0 = const()[name = tensor("op_2865_groups_0"), val = tensor(1)]; + tensor layers_7_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169469504))), name = tensor("layers_7_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169327040))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_2865_cast_fp16 = conv(dilations = var_2865_dilations_0, groups = var_2865_groups_0, pad = var_2865_pad_0, pad_type = var_2865_pad_type_0, strides = var_2865_strides_0, weight = layers_7_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_199_cast_fp16)[name = tensor("op_2865_cast_fp16")]; + tensor input_201_cast_fp16 = add(x = var_2859_cast_fp16, y = var_2865_cast_fp16)[name = tensor("input_201_cast_fp16")]; + tensor input_203_cast_fp16 = silu(x = input_201_cast_fp16)[name = tensor("input_203_cast_fp16")]; + tensor var_2876_pad_type_0 = const()[name = tensor("op_2876_pad_type_0"), val = tensor("valid")]; + tensor var_2876_strides_0 = const()[name = tensor("op_2876_strides_0"), val = tensor([1, 1])]; + tensor var_2876_pad_0 = const()[name = tensor("op_2876_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2876_dilations_0 = const()[name = tensor("op_2876_dilations_0"), val = tensor([1, 1])]; + tensor var_2876_groups_0 = const()[name = tensor("op_2876_groups_0"), val = tensor(1)]; + tensor layers_7_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169993856))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173139648))), name = tensor("layers_7_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_2876_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2876_dilations_0, groups = var_2876_groups_0, pad = var_2876_pad_0, pad_type = var_2876_pad_type_0, strides = var_2876_strides_0, weight = layers_7_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_203_cast_fp16)[name = tensor("op_2876_cast_fp16")]; + tensor var_2882_pad_type_0 = const()[name = tensor("op_2882_pad_type_0"), val = tensor("valid")]; + tensor var_2882_strides_0 = const()[name = tensor("op_2882_strides_0"), val = tensor([1, 1])]; + tensor var_2882_pad_0 = const()[name = tensor("op_2882_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2882_dilations_0 = const()[name = tensor("op_2882_dilations_0"), val = tensor([1, 1])]; + tensor var_2882_groups_0 = const()[name = tensor("op_2882_groups_0"), val = tensor(1)]; + tensor layers_7_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173311616))), name = tensor("layers_7_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173139840))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_2882_cast_fp16 = conv(dilations = var_2882_dilations_0, groups = var_2882_groups_0, pad = var_2882_pad_0, pad_type = var_2882_pad_type_0, strides = var_2882_strides_0, weight = layers_7_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_203_cast_fp16)[name = tensor("op_2882_cast_fp16")]; + tensor x_45_cast_fp16 = add(x = var_2876_cast_fp16, y = var_2882_cast_fp16)[name = tensor("x_45_cast_fp16")]; + tensor var_2884_to_fp16 = const()[name = tensor("op_2884_to_fp16"), val = tensor(0x1p-1)]; + tensor var_2885_cast_fp16 = mul(x = x_45_cast_fp16, y = var_2884_to_fp16)[name = tensor("op_2885_cast_fp16")]; + tensor inputs_73_cast_fp16 = add(x = inputs_71_cast_fp16, y = var_2885_cast_fp16)[name = tensor("inputs_73_cast_fp16")]; + tensor out_73_axes_0 = const()[name = tensor("out_73_axes_0"), val = tensor([1])]; + tensor var_2895_to_fp16 = const()[name = tensor("op_2895_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_73_cast_fp16 = layer_norm(axes = out_73_axes_0, epsilon = var_2895_to_fp16, x = inputs_73_cast_fp16)[name = tensor("out_73_cast_fp16")]; + tensor obj_31_gamma_0_to_fp16 = const()[name = tensor("obj_31_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173835968)))]; + tensor obj_31_beta_0_to_fp16 = const()[name = tensor("obj_31_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173838080)))]; + tensor obj_31_epsilon_0_to_fp16 = const()[name = tensor("obj_31_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_31_cast_fp16 = batch_norm(beta = obj_31_beta_0_to_fp16, epsilon = obj_31_epsilon_0_to_fp16, gamma = obj_31_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_73_cast_fp16)[name = tensor("obj_31_cast_fp16")]; + tensor var_2920_pad_type_0 = const()[name = tensor("op_2920_pad_type_0"), val = tensor("valid")]; + tensor var_2920_strides_0 = const()[name = tensor("op_2920_strides_0"), val = tensor([1, 1])]; + tensor var_2920_pad_0 = const()[name = tensor("op_2920_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2920_dilations_0 = const()[name = tensor("op_2920_dilations_0"), val = tensor([1, 1])]; + tensor var_2920_groups_0 = const()[name = tensor("op_2920_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(173840192))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174626688))), name = tensor("layers_7_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2920_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2920_dilations_0, groups = var_2920_groups_0, pad = var_2920_pad_0, pad_type = var_2920_pad_type_0, strides = var_2920_strides_0, weight = layers_7_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_31_cast_fp16)[name = tensor("op_2920_cast_fp16")]; + tensor var_2926_pad_type_0 = const()[name = tensor("op_2926_pad_type_0"), val = tensor("valid")]; + tensor var_2926_strides_0 = const()[name = tensor("op_2926_strides_0"), val = tensor([1, 1])]; + tensor var_2926_pad_0 = const()[name = tensor("op_2926_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2926_dilations_0 = const()[name = tensor("op_2926_dilations_0"), val = tensor([1, 1])]; + tensor var_2926_groups_0 = const()[name = tensor("op_2926_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174659584))), name = tensor("layers_7_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174626880))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2926_cast_fp16 = conv(dilations = var_2926_dilations_0, groups = var_2926_groups_0, pad = var_2926_pad_0, pad_type = var_2926_pad_type_0, strides = var_2926_strides_0, weight = layers_7_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_31_cast_fp16)[name = tensor("op_2926_cast_fp16")]; + tensor query_29_cast_fp16 = add(x = var_2920_cast_fp16, y = var_2926_cast_fp16)[name = tensor("query_29_cast_fp16")]; + tensor var_2935_pad_type_0 = const()[name = tensor("op_2935_pad_type_0"), val = tensor("valid")]; + tensor var_2935_strides_0 = const()[name = tensor("op_2935_strides_0"), val = tensor([1, 1])]; + tensor var_2935_pad_0 = const()[name = tensor("op_2935_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2935_dilations_0 = const()[name = tensor("op_2935_dilations_0"), val = tensor([1, 1])]; + tensor var_2935_groups_0 = const()[name = tensor("op_2935_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174790720))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175577216))), name = tensor("layers_7_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2935_cast_fp16 = conv(dilations = var_2935_dilations_0, groups = var_2935_groups_0, pad = var_2935_pad_0, pad_type = var_2935_pad_type_0, strides = var_2935_strides_0, weight = layers_7_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_31_cast_fp16)[name = tensor("op_2935_cast_fp16")]; + tensor var_2941_pad_type_0 = const()[name = tensor("op_2941_pad_type_0"), val = tensor("valid")]; + tensor var_2941_strides_0 = const()[name = tensor("op_2941_strides_0"), val = tensor([1, 1])]; + tensor var_2941_pad_0 = const()[name = tensor("op_2941_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2941_dilations_0 = const()[name = tensor("op_2941_dilations_0"), val = tensor([1, 1])]; + tensor var_2941_groups_0 = const()[name = tensor("op_2941_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175622784))), name = tensor("layers_7_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175577408))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2941_cast_fp16 = conv(dilations = var_2941_dilations_0, groups = var_2941_groups_0, pad = var_2941_pad_0, pad_type = var_2941_pad_type_0, strides = var_2941_strides_0, weight = layers_7_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_31_cast_fp16)[name = tensor("op_2941_cast_fp16")]; + tensor key_15_cast_fp16 = add(x = var_2935_cast_fp16, y = var_2941_cast_fp16)[name = tensor("key_15_cast_fp16")]; + tensor var_2951_pad_type_0 = const()[name = tensor("op_2951_pad_type_0"), val = tensor("valid")]; + tensor var_2951_strides_0 = const()[name = tensor("op_2951_strides_0"), val = tensor([1, 1])]; + tensor var_2951_pad_0 = const()[name = tensor("op_2951_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2951_dilations_0 = const()[name = tensor("op_2951_dilations_0"), val = tensor([1, 1])]; + tensor var_2951_groups_0 = const()[name = tensor("op_2951_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175753920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176540416))), name = tensor("layers_7_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2951_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2951_dilations_0, groups = var_2951_groups_0, pad = var_2951_pad_0, pad_type = var_2951_pad_type_0, strides = var_2951_strides_0, weight = layers_7_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_31_cast_fp16)[name = tensor("op_2951_cast_fp16")]; + tensor var_2957_pad_type_0 = const()[name = tensor("op_2957_pad_type_0"), val = tensor("valid")]; + tensor var_2957_strides_0 = const()[name = tensor("op_2957_strides_0"), val = tensor([1, 1])]; + tensor var_2957_pad_0 = const()[name = tensor("op_2957_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2957_dilations_0 = const()[name = tensor("op_2957_dilations_0"), val = tensor([1, 1])]; + tensor var_2957_groups_0 = const()[name = tensor("op_2957_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176577536))), name = tensor("layers_7_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176540608))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2957_cast_fp16 = conv(dilations = var_2957_dilations_0, groups = var_2957_groups_0, pad = var_2957_pad_0, pad_type = var_2957_pad_type_0, strides = var_2957_strides_0, weight = layers_7_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_31_cast_fp16)[name = tensor("op_2957_cast_fp16")]; + tensor value_15_cast_fp16 = add(x = var_2951_cast_fp16, y = var_2957_cast_fp16)[name = tensor("value_15_cast_fp16")]; + tensor var_2960_to_fp16 = const()[name = tensor("op_2960_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176708672)))]; + tensor query_31_cast_fp16 = add(x = query_29_cast_fp16, y = var_2960_to_fp16)[name = tensor("query_31_cast_fp16")]; + tensor var_2963_to_fp16 = const()[name = tensor("op_2963_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176710784)))]; + tensor q_with_bias_v_15_cast_fp16 = add(x = query_29_cast_fp16, y = var_2963_to_fp16)[name = tensor("q_with_bias_v_15_cast_fp16")]; + tensor var_2973_pad_type_0 = const()[name = tensor("op_2973_pad_type_0"), val = tensor("valid")]; + tensor var_2973_strides_0 = const()[name = tensor("op_2973_strides_0"), val = tensor([1, 1])]; + tensor var_2973_pad_0 = const()[name = tensor("op_2973_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2973_dilations_0 = const()[name = tensor("op_2973_dilations_0"), val = tensor([1, 1])]; + tensor var_2973_groups_0 = const()[name = tensor("op_2973_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176712896))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177499392))), name = tensor("layers_7_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2973_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2973_dilations_0, groups = var_2973_groups_0, pad = var_2973_pad_0, pad_type = var_2973_pad_type_0, strides = var_2973_strides_0, weight = layers_7_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_2973_cast_fp16")]; + tensor var_2979_pad_type_0 = const()[name = tensor("op_2979_pad_type_0"), val = tensor("valid")]; + tensor var_2979_strides_0 = const()[name = tensor("op_2979_strides_0"), val = tensor([1, 1])]; + tensor var_2979_pad_0 = const()[name = tensor("op_2979_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2979_dilations_0 = const()[name = tensor("op_2979_dilations_0"), val = tensor([1, 1])]; + tensor var_2979_groups_0 = const()[name = tensor("op_2979_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177573376))), name = tensor("layers_7_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177499584))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_2979_cast_fp16 = conv(dilations = var_2979_dilations_0, groups = var_2979_groups_0, pad = var_2979_pad_0, pad_type = var_2979_pad_type_0, strides = var_2979_strides_0, weight = layers_7_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_2979_cast_fp16")]; + tensor p_15_cast_fp16 = add(x = var_2973_cast_fp16, y = var_2979_cast_fp16)[name = tensor("p_15_cast_fp16")]; + tensor var_2983 = const()[name = tensor("op_2983"), val = tensor([1, 8, 128, 188])]; + tensor var_2984_cast_fp16 = reshape(shape = var_2983, x = q_with_bias_v_15_cast_fp16)[name = tensor("op_2984_cast_fp16")]; + tensor var_2985 = const()[name = tensor("op_2985"), val = tensor([1, 8, 128, -1])]; + tensor var_2986_cast_fp16 = reshape(shape = var_2985, x = p_15_cast_fp16)[name = tensor("op_2986_cast_fp16")]; + tensor matrix_bd_57_transpose_x_0 = const()[name = tensor("matrix_bd_57_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_57_transpose_y_0 = const()[name = tensor("matrix_bd_57_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_57_cast_fp16 = matmul(transpose_x = matrix_bd_57_transpose_x_0, transpose_y = matrix_bd_57_transpose_y_0, x = var_2984_cast_fp16, y = var_2986_cast_fp16)[name = tensor("matrix_bd_57_cast_fp16")]; + tensor matrix_bd_59_pad_0 = const()[name = tensor("matrix_bd_59_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_59_mode_0 = const()[name = tensor("matrix_bd_59_mode_0"), val = tensor("constant")]; + tensor const_87_to_fp16 = const()[name = tensor("const_87_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_59_cast_fp16 = pad(constant_val = const_87_to_fp16, mode = matrix_bd_59_mode_0, pad = matrix_bd_59_pad_0, x = matrix_bd_57_cast_fp16)[name = tensor("matrix_bd_59_cast_fp16")]; + tensor var_2995 = const()[name = tensor("op_2995"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_61_cast_fp16 = reshape(shape = var_2995, x = matrix_bd_59_cast_fp16)[name = tensor("matrix_bd_61_cast_fp16")]; + tensor var_2999_begin_0 = const()[name = tensor("op_2999_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_2999_end_0 = const()[name = tensor("op_2999_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_2999_end_mask_0 = const()[name = tensor("op_2999_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2999_cast_fp16 = slice_by_index(begin = var_2999_begin_0, end = var_2999_end_0, end_mask = var_2999_end_mask_0, x = matrix_bd_61_cast_fp16)[name = tensor("op_2999_cast_fp16")]; + tensor var_3000 = const()[name = tensor("op_3000"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_63_cast_fp16 = reshape(shape = var_3000, x = var_2999_cast_fp16)[name = tensor("matrix_bd_63_cast_fp16")]; + tensor var_3005_begin_0 = const()[name = tensor("op_3005_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3005_end_0 = const()[name = tensor("op_3005_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_3005_end_mask_0 = const()[name = tensor("op_3005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3005_cast_fp16 = slice_by_index(begin = var_3005_begin_0, end = var_3005_end_0, end_mask = var_3005_end_mask_0, x = matrix_bd_63_cast_fp16)[name = tensor("op_3005_cast_fp16")]; + tensor var_3006_to_fp16 = const()[name = tensor("op_3006_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_15_cast_fp16 = mul(x = var_3005_cast_fp16, y = var_3006_to_fp16)[name = tensor("qk_mask_15_cast_fp16")]; + tensor var_3010 = const()[name = tensor("op_3010"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_15_cast_fp16 = reshape(shape = var_3010, x = query_31_cast_fp16)[name = tensor("mh_q_15_cast_fp16")]; + tensor var_3012_to_fp16 = const()[name = tensor("op_3012_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_3013_cast_fp16 = mul(x = mh_q_15_cast_fp16, y = var_3012_to_fp16)[name = tensor("op_3013_cast_fp16")]; + tensor var_3016 = const()[name = tensor("op_3016"), val = tensor([1, 8, 128, 188])]; + tensor var_3017_cast_fp16 = reshape(shape = var_3016, x = key_15_cast_fp16)[name = tensor("op_3017_cast_fp16")]; + tensor mh_w_29_transpose_x_0 = const()[name = tensor("mh_w_29_transpose_x_0"), val = tensor(true)]; + tensor mh_w_29_transpose_y_0 = const()[name = tensor("mh_w_29_transpose_y_0"), val = tensor(false)]; + tensor mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_3013_cast_fp16, y = var_3017_cast_fp16)[name = tensor("mh_w_29_cast_fp16")]; + tensor mh_w_31_cast_fp16 = add(x = mh_w_29_cast_fp16, y = qk_mask_15_cast_fp16)[name = tensor("mh_w_31_cast_fp16")]; + tensor var_3021_cast_fp16 = softmax(axis = var_2808, x = mh_w_31_cast_fp16)[name = tensor("op_3021_cast_fp16")]; + tensor var_3022 = const()[name = tensor("op_3022"), val = tensor([1, 8, 128, 188])]; + tensor var_3023_cast_fp16 = reshape(shape = var_3022, x = value_15_cast_fp16)[name = tensor("op_3023_cast_fp16")]; + tensor attn_15_transpose_x_0 = const()[name = tensor("attn_15_transpose_x_0"), val = tensor(false)]; + tensor attn_15_transpose_y_0 = const()[name = tensor("attn_15_transpose_y_0"), val = tensor(true)]; + tensor attn_15_cast_fp16 = matmul(transpose_x = attn_15_transpose_x_0, transpose_y = attn_15_transpose_y_0, x = var_3023_cast_fp16, y = var_3021_cast_fp16)[name = tensor("attn_15_cast_fp16")]; + tensor var_3026 = const()[name = tensor("op_3026"), val = tensor([1, 1024, 1, 188])]; + tensor input_205_cast_fp16 = reshape(shape = var_3026, x = attn_15_cast_fp16)[name = tensor("input_205_cast_fp16")]; + tensor var_3036_pad_type_0 = const()[name = tensor("op_3036_pad_type_0"), val = tensor("valid")]; + tensor var_3036_strides_0 = const()[name = tensor("op_3036_strides_0"), val = tensor([1, 1])]; + tensor var_3036_pad_0 = const()[name = tensor("op_3036_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3036_dilations_0 = const()[name = tensor("op_3036_dilations_0"), val = tensor([1, 1])]; + tensor var_3036_groups_0 = const()[name = tensor("op_3036_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177704512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178491008))), name = tensor("layers_7_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3036_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3036_dilations_0, groups = var_3036_groups_0, pad = var_3036_pad_0, pad_type = var_3036_pad_type_0, strides = var_3036_strides_0, weight = layers_7_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_205_cast_fp16)[name = tensor("op_3036_cast_fp16")]; + tensor var_3042_pad_type_0 = const()[name = tensor("op_3042_pad_type_0"), val = tensor("valid")]; + tensor var_3042_strides_0 = const()[name = tensor("op_3042_strides_0"), val = tensor([1, 1])]; + tensor var_3042_pad_0 = const()[name = tensor("op_3042_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3042_dilations_0 = const()[name = tensor("op_3042_dilations_0"), val = tensor([1, 1])]; + tensor var_3042_groups_0 = const()[name = tensor("op_3042_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178520640))), name = tensor("layers_7_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178491200))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3042_cast_fp16 = conv(dilations = var_3042_dilations_0, groups = var_3042_groups_0, pad = var_3042_pad_0, pad_type = var_3042_pad_type_0, strides = var_3042_strides_0, weight = layers_7_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_205_cast_fp16)[name = tensor("op_3042_cast_fp16")]; + tensor obj_33_cast_fp16 = add(x = var_3036_cast_fp16, y = var_3042_cast_fp16)[name = tensor("obj_33_cast_fp16")]; + tensor inputs_75_cast_fp16 = add(x = inputs_73_cast_fp16, y = obj_33_cast_fp16)[name = tensor("inputs_75_cast_fp16")]; + tensor out_75_axes_0 = const()[name = tensor("out_75_axes_0"), val = tensor([1])]; + tensor var_3053_to_fp16 = const()[name = tensor("op_3053_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_75_cast_fp16 = layer_norm(axes = out_75_axes_0, epsilon = var_3053_to_fp16, x = inputs_75_cast_fp16)[name = tensor("out_75_cast_fp16")]; + tensor input_207_gamma_0_to_fp16 = const()[name = tensor("input_207_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178651776)))]; + tensor input_207_beta_0_to_fp16 = const()[name = tensor("input_207_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178653888)))]; + tensor input_207_epsilon_0_to_fp16 = const()[name = tensor("input_207_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_207_cast_fp16 = batch_norm(beta = input_207_beta_0_to_fp16, epsilon = input_207_epsilon_0_to_fp16, gamma = input_207_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_75_cast_fp16)[name = tensor("input_207_cast_fp16")]; + tensor var_3074_pad_type_0 = const()[name = tensor("op_3074_pad_type_0"), val = tensor("valid")]; + tensor var_3074_strides_0 = const()[name = tensor("op_3074_strides_0"), val = tensor([1, 1])]; + tensor var_3074_pad_0 = const()[name = tensor("op_3074_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3074_dilations_0 = const()[name = tensor("op_3074_dilations_0"), val = tensor([1, 1])]; + tensor var_3074_groups_0 = const()[name = tensor("op_3074_groups_0"), val = tensor(1)]; + tensor layers_7_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178656000))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(180228928))), name = tensor("layers_7_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_3074_cast_fp16 = conv(dilations = var_3074_dilations_0, groups = var_3074_groups_0, pad = var_3074_pad_0, pad_type = var_3074_pad_type_0, strides = var_3074_strides_0, weight = layers_7_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_207_cast_fp16)[name = tensor("op_3074_cast_fp16")]; + tensor var_3080_pad_type_0 = const()[name = tensor("op_3080_pad_type_0"), val = tensor("valid")]; + tensor var_3080_strides_0 = const()[name = tensor("op_3080_strides_0"), val = tensor([1, 1])]; + tensor var_3080_pad_0 = const()[name = tensor("op_3080_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3080_dilations_0 = const()[name = tensor("op_3080_dilations_0"), val = tensor([1, 1])]; + tensor var_3080_groups_0 = const()[name = tensor("op_3080_groups_0"), val = tensor(1)]; + tensor layers_7_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(180292672))), name = tensor("layers_7_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(180229120))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_3080_cast_fp16 = conv(dilations = var_3080_dilations_0, groups = var_3080_groups_0, pad = var_3080_pad_0, pad_type = var_3080_pad_type_0, strides = var_3080_strides_0, weight = layers_7_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_207_cast_fp16)[name = tensor("op_3080_cast_fp16")]; + tensor input_209_cast_fp16 = add(x = var_3074_cast_fp16, y = var_3080_cast_fp16)[name = tensor("input_209_cast_fp16")]; + tensor input_211_split_num_splits_0 = const()[name = tensor("input_211_split_num_splits_0"), val = tensor(2)]; + tensor input_211_split_axis_0 = const()[name = tensor("input_211_split_axis_0"), val = tensor(1)]; + tensor input_211_split_cast_fp16_0, tensor input_211_split_cast_fp16_1 = split(axis = input_211_split_axis_0, num_splits = input_211_split_num_splits_0, x = input_209_cast_fp16)[name = tensor("input_211_split_cast_fp16")]; + tensor input_211_split_1_sigmoid_cast_fp16 = sigmoid(x = input_211_split_cast_fp16_1)[name = tensor("input_211_split_1_sigmoid_cast_fp16")]; + tensor input_211_cast_fp16 = mul(x = input_211_split_cast_fp16_0, y = input_211_split_1_sigmoid_cast_fp16)[name = tensor("input_211_cast_fp16")]; + tensor input_213_pad_type_0 = const()[name = tensor("input_213_pad_type_0"), val = tensor("custom")]; + tensor input_213_pad_0 = const()[name = tensor("input_213_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_213_groups_0 = const()[name = tensor("input_213_groups_0"), val = tensor(1024)]; + tensor input_213_strides_0 = const()[name = tensor("input_213_strides_0"), val = tensor([1, 1])]; + tensor input_213_dilations_0 = const()[name = tensor("input_213_dilations_0"), val = tensor([1, 1])]; + tensor const_282_to_fp16 = const()[name = tensor("const_282_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(180554880)))]; + tensor const_283_to_fp16 = const()[name = tensor("const_283_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(180573376)))]; + tensor input_215_cast_fp16 = conv(bias = const_283_to_fp16, dilations = input_213_dilations_0, groups = input_213_groups_0, pad = input_213_pad_0, pad_type = input_213_pad_type_0, strides = input_213_strides_0, weight = const_282_to_fp16, x = input_211_cast_fp16)[name = tensor("input_215_cast_fp16")]; + tensor input_217_cast_fp16 = silu(x = input_215_cast_fp16)[name = tensor("input_217_cast_fp16")]; + tensor var_3102_pad_type_0 = const()[name = tensor("op_3102_pad_type_0"), val = tensor("valid")]; + tensor var_3102_strides_0 = const()[name = tensor("op_3102_strides_0"), val = tensor([1, 1])]; + tensor var_3102_pad_0 = const()[name = tensor("op_3102_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3102_dilations_0 = const()[name = tensor("op_3102_dilations_0"), val = tensor([1, 1])]; + tensor var_3102_groups_0 = const()[name = tensor("op_3102_groups_0"), val = tensor(1)]; + tensor layers_7_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(180575488))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181361984))), name = tensor("layers_7_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3102_cast_fp16 = conv(dilations = var_3102_dilations_0, groups = var_3102_groups_0, pad = var_3102_pad_0, pad_type = var_3102_pad_type_0, strides = var_3102_strides_0, weight = layers_7_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_217_cast_fp16)[name = tensor("op_3102_cast_fp16")]; + tensor var_3108_pad_type_0 = const()[name = tensor("op_3108_pad_type_0"), val = tensor("valid")]; + tensor var_3108_strides_0 = const()[name = tensor("op_3108_strides_0"), val = tensor([1, 1])]; + tensor var_3108_pad_0 = const()[name = tensor("op_3108_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3108_dilations_0 = const()[name = tensor("op_3108_dilations_0"), val = tensor([1, 1])]; + tensor var_3108_groups_0 = const()[name = tensor("op_3108_groups_0"), val = tensor(1)]; + tensor layers_7_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181393472))), name = tensor("layers_7_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181362176))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3108_cast_fp16 = conv(dilations = var_3108_dilations_0, groups = var_3108_groups_0, pad = var_3108_pad_0, pad_type = var_3108_pad_type_0, strides = var_3108_strides_0, weight = layers_7_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_217_cast_fp16)[name = tensor("op_3108_cast_fp16")]; + tensor x_47_cast_fp16 = add(x = var_3102_cast_fp16, y = var_3108_cast_fp16)[name = tensor("x_47_cast_fp16")]; + tensor inputs_77_cast_fp16 = add(x = inputs_75_cast_fp16, y = x_47_cast_fp16)[name = tensor("inputs_77_cast_fp16")]; + tensor out_77_axes_0 = const()[name = tensor("out_77_axes_0"), val = tensor([1])]; + tensor var_3119_to_fp16 = const()[name = tensor("op_3119_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_77_cast_fp16 = layer_norm(axes = out_77_axes_0, epsilon = var_3119_to_fp16, x = inputs_77_cast_fp16)[name = tensor("out_77_cast_fp16")]; + tensor input_219_gamma_0_to_fp16 = const()[name = tensor("input_219_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181524608)))]; + tensor input_219_beta_0_to_fp16 = const()[name = tensor("input_219_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181526720)))]; + tensor input_219_epsilon_0_to_fp16 = const()[name = tensor("input_219_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_219_cast_fp16 = batch_norm(beta = input_219_beta_0_to_fp16, epsilon = input_219_epsilon_0_to_fp16, gamma = input_219_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_77_cast_fp16)[name = tensor("input_219_cast_fp16")]; + tensor var_3139_pad_type_0 = const()[name = tensor("op_3139_pad_type_0"), val = tensor("valid")]; + tensor var_3139_strides_0 = const()[name = tensor("op_3139_strides_0"), val = tensor([1, 1])]; + tensor var_3139_pad_0 = const()[name = tensor("op_3139_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3139_dilations_0 = const()[name = tensor("op_3139_dilations_0"), val = tensor([1, 1])]; + tensor var_3139_groups_0 = const()[name = tensor("op_3139_groups_0"), val = tensor(1)]; + tensor layers_7_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181528832))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184674624))), name = tensor("layers_7_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_3139_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_3139_dilations_0, groups = var_3139_groups_0, pad = var_3139_pad_0, pad_type = var_3139_pad_type_0, strides = var_3139_strides_0, weight = layers_7_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_219_cast_fp16)[name = tensor("op_3139_cast_fp16")]; + tensor var_3145_pad_type_0 = const()[name = tensor("op_3145_pad_type_0"), val = tensor("valid")]; + tensor var_3145_strides_0 = const()[name = tensor("op_3145_strides_0"), val = tensor([1, 1])]; + tensor var_3145_pad_0 = const()[name = tensor("op_3145_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3145_dilations_0 = const()[name = tensor("op_3145_dilations_0"), val = tensor([1, 1])]; + tensor var_3145_groups_0 = const()[name = tensor("op_3145_groups_0"), val = tensor(1)]; + tensor layers_7_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184823616))), name = tensor("layers_7_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184674816))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_3145_cast_fp16 = conv(dilations = var_3145_dilations_0, groups = var_3145_groups_0, pad = var_3145_pad_0, pad_type = var_3145_pad_type_0, strides = var_3145_strides_0, weight = layers_7_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_219_cast_fp16)[name = tensor("op_3145_cast_fp16")]; + tensor input_221_cast_fp16 = add(x = var_3139_cast_fp16, y = var_3145_cast_fp16)[name = tensor("input_221_cast_fp16")]; + tensor input_223_cast_fp16 = silu(x = input_221_cast_fp16)[name = tensor("input_223_cast_fp16")]; + tensor var_3156_pad_type_0 = const()[name = tensor("op_3156_pad_type_0"), val = tensor("valid")]; + tensor var_3156_strides_0 = const()[name = tensor("op_3156_strides_0"), val = tensor([1, 1])]; + tensor var_3156_pad_0 = const()[name = tensor("op_3156_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3156_dilations_0 = const()[name = tensor("op_3156_dilations_0"), val = tensor([1, 1])]; + tensor var_3156_groups_0 = const()[name = tensor("op_3156_groups_0"), val = tensor(1)]; + tensor layers_7_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(185347968))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(188493760))), name = tensor("layers_7_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_3156_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3156_dilations_0, groups = var_3156_groups_0, pad = var_3156_pad_0, pad_type = var_3156_pad_type_0, strides = var_3156_strides_0, weight = layers_7_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_223_cast_fp16)[name = tensor("op_3156_cast_fp16")]; + tensor var_3162_pad_type_0 = const()[name = tensor("op_3162_pad_type_0"), val = tensor("valid")]; + tensor var_3162_strides_0 = const()[name = tensor("op_3162_strides_0"), val = tensor([1, 1])]; + tensor var_3162_pad_0 = const()[name = tensor("op_3162_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3162_dilations_0 = const()[name = tensor("op_3162_dilations_0"), val = tensor([1, 1])]; + tensor var_3162_groups_0 = const()[name = tensor("op_3162_groups_0"), val = tensor(1)]; + tensor layers_7_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(188644160))), name = tensor("layers_7_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(188493952))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_3162_cast_fp16 = conv(dilations = var_3162_dilations_0, groups = var_3162_groups_0, pad = var_3162_pad_0, pad_type = var_3162_pad_type_0, strides = var_3162_strides_0, weight = layers_7_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_223_cast_fp16)[name = tensor("op_3162_cast_fp16")]; + tensor x_49_cast_fp16 = add(x = var_3156_cast_fp16, y = var_3162_cast_fp16)[name = tensor("x_49_cast_fp16")]; + tensor var_3164_to_fp16 = const()[name = tensor("op_3164_to_fp16"), val = tensor(0x1p-1)]; + tensor var_3165_cast_fp16 = mul(x = x_49_cast_fp16, y = var_3164_to_fp16)[name = tensor("op_3165_cast_fp16")]; + tensor inputs_79_cast_fp16 = add(x = inputs_77_cast_fp16, y = var_3165_cast_fp16)[name = tensor("inputs_79_cast_fp16")]; + tensor out_79_axes_0 = const()[name = tensor("out_79_axes_0"), val = tensor([1])]; + tensor var_3175_to_fp16 = const()[name = tensor("op_3175_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_79_cast_fp16 = layer_norm(axes = out_79_axes_0, epsilon = var_3175_to_fp16, x = inputs_79_cast_fp16)[name = tensor("out_79_cast_fp16")]; + tensor inputs_81_gamma_0_to_fp16 = const()[name = tensor("inputs_81_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189168512)))]; + tensor inputs_81_beta_0_to_fp16 = const()[name = tensor("inputs_81_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189170624)))]; + tensor inputs_81_epsilon_0_to_fp16 = const()[name = tensor("inputs_81_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_81_cast_fp16 = batch_norm(beta = inputs_81_beta_0_to_fp16, epsilon = inputs_81_epsilon_0_to_fp16, gamma = inputs_81_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_79_cast_fp16)[name = tensor("inputs_81_cast_fp16")]; + tensor var_3189 = const()[name = tensor("op_3189"), val = tensor(3)]; + tensor out_81_axes_0 = const()[name = tensor("out_81_axes_0"), val = tensor([1])]; + tensor var_3220_to_fp16 = const()[name = tensor("op_3220_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_81_cast_fp16 = layer_norm(axes = out_81_axes_0, epsilon = var_3220_to_fp16, x = inputs_81_cast_fp16)[name = tensor("out_81_cast_fp16")]; + tensor input_225_gamma_0_to_fp16 = const()[name = tensor("input_225_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189172736)))]; + tensor input_225_beta_0_to_fp16 = const()[name = tensor("input_225_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189174848)))]; + tensor input_225_epsilon_0_to_fp16 = const()[name = tensor("input_225_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_225_cast_fp16 = batch_norm(beta = input_225_beta_0_to_fp16, epsilon = input_225_epsilon_0_to_fp16, gamma = input_225_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_81_cast_fp16)[name = tensor("input_225_cast_fp16")]; + tensor var_3240_pad_type_0 = const()[name = tensor("op_3240_pad_type_0"), val = tensor("valid")]; + tensor var_3240_strides_0 = const()[name = tensor("op_3240_strides_0"), val = tensor([1, 1])]; + tensor var_3240_pad_0 = const()[name = tensor("op_3240_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3240_dilations_0 = const()[name = tensor("op_3240_dilations_0"), val = tensor([1, 1])]; + tensor var_3240_groups_0 = const()[name = tensor("op_3240_groups_0"), val = tensor(1)]; + tensor layers_8_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(189176960))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(192322752))), name = tensor("layers_8_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_3240_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_3240_dilations_0, groups = var_3240_groups_0, pad = var_3240_pad_0, pad_type = var_3240_pad_type_0, strides = var_3240_strides_0, weight = layers_8_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_225_cast_fp16)[name = tensor("op_3240_cast_fp16")]; + tensor var_3246_pad_type_0 = const()[name = tensor("op_3246_pad_type_0"), val = tensor("valid")]; + tensor var_3246_strides_0 = const()[name = tensor("op_3246_strides_0"), val = tensor([1, 1])]; + tensor var_3246_pad_0 = const()[name = tensor("op_3246_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3246_dilations_0 = const()[name = tensor("op_3246_dilations_0"), val = tensor([1, 1])]; + tensor var_3246_groups_0 = const()[name = tensor("op_3246_groups_0"), val = tensor(1)]; + tensor layers_8_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(192468352))), name = tensor("layers_8_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(192322944))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_3246_cast_fp16 = conv(dilations = var_3246_dilations_0, groups = var_3246_groups_0, pad = var_3246_pad_0, pad_type = var_3246_pad_type_0, strides = var_3246_strides_0, weight = layers_8_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_225_cast_fp16)[name = tensor("op_3246_cast_fp16")]; + tensor input_227_cast_fp16 = add(x = var_3240_cast_fp16, y = var_3246_cast_fp16)[name = tensor("input_227_cast_fp16")]; + tensor input_229_cast_fp16 = silu(x = input_227_cast_fp16)[name = tensor("input_229_cast_fp16")]; + tensor var_3257_pad_type_0 = const()[name = tensor("op_3257_pad_type_0"), val = tensor("valid")]; + tensor var_3257_strides_0 = const()[name = tensor("op_3257_strides_0"), val = tensor([1, 1])]; + tensor var_3257_pad_0 = const()[name = tensor("op_3257_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3257_dilations_0 = const()[name = tensor("op_3257_dilations_0"), val = tensor([1, 1])]; + tensor var_3257_groups_0 = const()[name = tensor("op_3257_groups_0"), val = tensor(1)]; + tensor layers_8_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(192992704))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196138496))), name = tensor("layers_8_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_3257_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3257_dilations_0, groups = var_3257_groups_0, pad = var_3257_pad_0, pad_type = var_3257_pad_type_0, strides = var_3257_strides_0, weight = layers_8_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_229_cast_fp16)[name = tensor("op_3257_cast_fp16")]; + tensor var_3263_pad_type_0 = const()[name = tensor("op_3263_pad_type_0"), val = tensor("valid")]; + tensor var_3263_strides_0 = const()[name = tensor("op_3263_strides_0"), val = tensor([1, 1])]; + tensor var_3263_pad_0 = const()[name = tensor("op_3263_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3263_dilations_0 = const()[name = tensor("op_3263_dilations_0"), val = tensor([1, 1])]; + tensor var_3263_groups_0 = const()[name = tensor("op_3263_groups_0"), val = tensor(1)]; + tensor layers_8_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196302400))), name = tensor("layers_8_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196138688))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_3263_cast_fp16 = conv(dilations = var_3263_dilations_0, groups = var_3263_groups_0, pad = var_3263_pad_0, pad_type = var_3263_pad_type_0, strides = var_3263_strides_0, weight = layers_8_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_229_cast_fp16)[name = tensor("op_3263_cast_fp16")]; + tensor x_51_cast_fp16 = add(x = var_3257_cast_fp16, y = var_3263_cast_fp16)[name = tensor("x_51_cast_fp16")]; + tensor var_3265_to_fp16 = const()[name = tensor("op_3265_to_fp16"), val = tensor(0x1p-1)]; + tensor var_3266_cast_fp16 = mul(x = x_51_cast_fp16, y = var_3265_to_fp16)[name = tensor("op_3266_cast_fp16")]; + tensor inputs_83_cast_fp16 = add(x = inputs_81_cast_fp16, y = var_3266_cast_fp16)[name = tensor("inputs_83_cast_fp16")]; + tensor out_83_axes_0 = const()[name = tensor("out_83_axes_0"), val = tensor([1])]; + tensor var_3276_to_fp16 = const()[name = tensor("op_3276_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_83_cast_fp16 = layer_norm(axes = out_83_axes_0, epsilon = var_3276_to_fp16, x = inputs_83_cast_fp16)[name = tensor("out_83_cast_fp16")]; + tensor obj_35_gamma_0_to_fp16 = const()[name = tensor("obj_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196826752)))]; + tensor obj_35_beta_0_to_fp16 = const()[name = tensor("obj_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196828864)))]; + tensor obj_35_epsilon_0_to_fp16 = const()[name = tensor("obj_35_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_35_cast_fp16 = batch_norm(beta = obj_35_beta_0_to_fp16, epsilon = obj_35_epsilon_0_to_fp16, gamma = obj_35_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_83_cast_fp16)[name = tensor("obj_35_cast_fp16")]; + tensor var_3301_pad_type_0 = const()[name = tensor("op_3301_pad_type_0"), val = tensor("valid")]; + tensor var_3301_strides_0 = const()[name = tensor("op_3301_strides_0"), val = tensor([1, 1])]; + tensor var_3301_pad_0 = const()[name = tensor("op_3301_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3301_dilations_0 = const()[name = tensor("op_3301_dilations_0"), val = tensor([1, 1])]; + tensor var_3301_groups_0 = const()[name = tensor("op_3301_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(196830976))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197617472))), name = tensor("layers_8_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3301_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3301_dilations_0, groups = var_3301_groups_0, pad = var_3301_pad_0, pad_type = var_3301_pad_type_0, strides = var_3301_strides_0, weight = layers_8_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_35_cast_fp16)[name = tensor("op_3301_cast_fp16")]; + tensor var_3307_pad_type_0 = const()[name = tensor("op_3307_pad_type_0"), val = tensor("valid")]; + tensor var_3307_strides_0 = const()[name = tensor("op_3307_strides_0"), val = tensor([1, 1])]; + tensor var_3307_pad_0 = const()[name = tensor("op_3307_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3307_dilations_0 = const()[name = tensor("op_3307_dilations_0"), val = tensor([1, 1])]; + tensor var_3307_groups_0 = const()[name = tensor("op_3307_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197654720))), name = tensor("layers_8_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197617664))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3307_cast_fp16 = conv(dilations = var_3307_dilations_0, groups = var_3307_groups_0, pad = var_3307_pad_0, pad_type = var_3307_pad_type_0, strides = var_3307_strides_0, weight = layers_8_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_35_cast_fp16)[name = tensor("op_3307_cast_fp16")]; + tensor query_33_cast_fp16 = add(x = var_3301_cast_fp16, y = var_3307_cast_fp16)[name = tensor("query_33_cast_fp16")]; + tensor var_3316_pad_type_0 = const()[name = tensor("op_3316_pad_type_0"), val = tensor("valid")]; + tensor var_3316_strides_0 = const()[name = tensor("op_3316_strides_0"), val = tensor([1, 1])]; + tensor var_3316_pad_0 = const()[name = tensor("op_3316_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3316_dilations_0 = const()[name = tensor("op_3316_dilations_0"), val = tensor([1, 1])]; + tensor var_3316_groups_0 = const()[name = tensor("op_3316_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197785856))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198572352))), name = tensor("layers_8_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3316_cast_fp16 = conv(dilations = var_3316_dilations_0, groups = var_3316_groups_0, pad = var_3316_pad_0, pad_type = var_3316_pad_type_0, strides = var_3316_strides_0, weight = layers_8_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_35_cast_fp16)[name = tensor("op_3316_cast_fp16")]; + tensor var_3322_pad_type_0 = const()[name = tensor("op_3322_pad_type_0"), val = tensor("valid")]; + tensor var_3322_strides_0 = const()[name = tensor("op_3322_strides_0"), val = tensor([1, 1])]; + tensor var_3322_pad_0 = const()[name = tensor("op_3322_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3322_dilations_0 = const()[name = tensor("op_3322_dilations_0"), val = tensor([1, 1])]; + tensor var_3322_groups_0 = const()[name = tensor("op_3322_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198621824))), name = tensor("layers_8_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198572544))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3322_cast_fp16 = conv(dilations = var_3322_dilations_0, groups = var_3322_groups_0, pad = var_3322_pad_0, pad_type = var_3322_pad_type_0, strides = var_3322_strides_0, weight = layers_8_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_35_cast_fp16)[name = tensor("op_3322_cast_fp16")]; + tensor key_17_cast_fp16 = add(x = var_3316_cast_fp16, y = var_3322_cast_fp16)[name = tensor("key_17_cast_fp16")]; + tensor var_3332_pad_type_0 = const()[name = tensor("op_3332_pad_type_0"), val = tensor("valid")]; + tensor var_3332_strides_0 = const()[name = tensor("op_3332_strides_0"), val = tensor([1, 1])]; + tensor var_3332_pad_0 = const()[name = tensor("op_3332_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3332_dilations_0 = const()[name = tensor("op_3332_dilations_0"), val = tensor([1, 1])]; + tensor var_3332_groups_0 = const()[name = tensor("op_3332_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198752960))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199539456))), name = tensor("layers_8_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3332_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3332_dilations_0, groups = var_3332_groups_0, pad = var_3332_pad_0, pad_type = var_3332_pad_type_0, strides = var_3332_strides_0, weight = layers_8_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_35_cast_fp16)[name = tensor("op_3332_cast_fp16")]; + tensor var_3338_pad_type_0 = const()[name = tensor("op_3338_pad_type_0"), val = tensor("valid")]; + tensor var_3338_strides_0 = const()[name = tensor("op_3338_strides_0"), val = tensor([1, 1])]; + tensor var_3338_pad_0 = const()[name = tensor("op_3338_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3338_dilations_0 = const()[name = tensor("op_3338_dilations_0"), val = tensor([1, 1])]; + tensor var_3338_groups_0 = const()[name = tensor("op_3338_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199577024))), name = tensor("layers_8_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199539648))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3338_cast_fp16 = conv(dilations = var_3338_dilations_0, groups = var_3338_groups_0, pad = var_3338_pad_0, pad_type = var_3338_pad_type_0, strides = var_3338_strides_0, weight = layers_8_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_35_cast_fp16)[name = tensor("op_3338_cast_fp16")]; + tensor value_17_cast_fp16 = add(x = var_3332_cast_fp16, y = var_3338_cast_fp16)[name = tensor("value_17_cast_fp16")]; + tensor var_3341_to_fp16 = const()[name = tensor("op_3341_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199708160)))]; + tensor query_35_cast_fp16 = add(x = query_33_cast_fp16, y = var_3341_to_fp16)[name = tensor("query_35_cast_fp16")]; + tensor var_3344_to_fp16 = const()[name = tensor("op_3344_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199710272)))]; + tensor q_with_bias_v_17_cast_fp16 = add(x = query_33_cast_fp16, y = var_3344_to_fp16)[name = tensor("q_with_bias_v_17_cast_fp16")]; + tensor var_3354_pad_type_0 = const()[name = tensor("op_3354_pad_type_0"), val = tensor("valid")]; + tensor var_3354_strides_0 = const()[name = tensor("op_3354_strides_0"), val = tensor([1, 1])]; + tensor var_3354_pad_0 = const()[name = tensor("op_3354_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3354_dilations_0 = const()[name = tensor("op_3354_dilations_0"), val = tensor([1, 1])]; + tensor var_3354_groups_0 = const()[name = tensor("op_3354_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(199712384))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(200498880))), name = tensor("layers_8_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3354_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3354_dilations_0, groups = var_3354_groups_0, pad = var_3354_pad_0, pad_type = var_3354_pad_type_0, strides = var_3354_strides_0, weight = layers_8_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_3354_cast_fp16")]; + tensor var_3360_pad_type_0 = const()[name = tensor("op_3360_pad_type_0"), val = tensor("valid")]; + tensor var_3360_strides_0 = const()[name = tensor("op_3360_strides_0"), val = tensor([1, 1])]; + tensor var_3360_pad_0 = const()[name = tensor("op_3360_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3360_dilations_0 = const()[name = tensor("op_3360_dilations_0"), val = tensor([1, 1])]; + tensor var_3360_groups_0 = const()[name = tensor("op_3360_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(200572224))), name = tensor("layers_8_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(200499072))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3360_cast_fp16 = conv(dilations = var_3360_dilations_0, groups = var_3360_groups_0, pad = var_3360_pad_0, pad_type = var_3360_pad_type_0, strides = var_3360_strides_0, weight = layers_8_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_3360_cast_fp16")]; + tensor p_17_cast_fp16 = add(x = var_3354_cast_fp16, y = var_3360_cast_fp16)[name = tensor("p_17_cast_fp16")]; + tensor var_3364 = const()[name = tensor("op_3364"), val = tensor([1, 8, 128, 188])]; + tensor var_3365_cast_fp16 = reshape(shape = var_3364, x = q_with_bias_v_17_cast_fp16)[name = tensor("op_3365_cast_fp16")]; + tensor var_3366 = const()[name = tensor("op_3366"), val = tensor([1, 8, 128, -1])]; + tensor var_3367_cast_fp16 = reshape(shape = var_3366, x = p_17_cast_fp16)[name = tensor("op_3367_cast_fp16")]; + tensor matrix_bd_65_transpose_x_0 = const()[name = tensor("matrix_bd_65_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_65_transpose_y_0 = const()[name = tensor("matrix_bd_65_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_65_cast_fp16 = matmul(transpose_x = matrix_bd_65_transpose_x_0, transpose_y = matrix_bd_65_transpose_y_0, x = var_3365_cast_fp16, y = var_3367_cast_fp16)[name = tensor("matrix_bd_65_cast_fp16")]; + tensor matrix_bd_67_pad_0 = const()[name = tensor("matrix_bd_67_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_67_mode_0 = const()[name = tensor("matrix_bd_67_mode_0"), val = tensor("constant")]; + tensor const_98_to_fp16 = const()[name = tensor("const_98_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_67_cast_fp16 = pad(constant_val = const_98_to_fp16, mode = matrix_bd_67_mode_0, pad = matrix_bd_67_pad_0, x = matrix_bd_65_cast_fp16)[name = tensor("matrix_bd_67_cast_fp16")]; + tensor var_3376 = const()[name = tensor("op_3376"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_69_cast_fp16 = reshape(shape = var_3376, x = matrix_bd_67_cast_fp16)[name = tensor("matrix_bd_69_cast_fp16")]; + tensor var_3380_begin_0 = const()[name = tensor("op_3380_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_3380_end_0 = const()[name = tensor("op_3380_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_3380_end_mask_0 = const()[name = tensor("op_3380_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3380_cast_fp16 = slice_by_index(begin = var_3380_begin_0, end = var_3380_end_0, end_mask = var_3380_end_mask_0, x = matrix_bd_69_cast_fp16)[name = tensor("op_3380_cast_fp16")]; + tensor var_3381 = const()[name = tensor("op_3381"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_71_cast_fp16 = reshape(shape = var_3381, x = var_3380_cast_fp16)[name = tensor("matrix_bd_71_cast_fp16")]; + tensor var_3386_begin_0 = const()[name = tensor("op_3386_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3386_end_0 = const()[name = tensor("op_3386_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_3386_end_mask_0 = const()[name = tensor("op_3386_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3386_cast_fp16 = slice_by_index(begin = var_3386_begin_0, end = var_3386_end_0, end_mask = var_3386_end_mask_0, x = matrix_bd_71_cast_fp16)[name = tensor("op_3386_cast_fp16")]; + tensor var_3387_to_fp16 = const()[name = tensor("op_3387_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_17_cast_fp16 = mul(x = var_3386_cast_fp16, y = var_3387_to_fp16)[name = tensor("qk_mask_17_cast_fp16")]; + tensor var_3391 = const()[name = tensor("op_3391"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_17_cast_fp16 = reshape(shape = var_3391, x = query_35_cast_fp16)[name = tensor("mh_q_17_cast_fp16")]; + tensor var_3393_to_fp16 = const()[name = tensor("op_3393_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_3394_cast_fp16 = mul(x = mh_q_17_cast_fp16, y = var_3393_to_fp16)[name = tensor("op_3394_cast_fp16")]; + tensor var_3397 = const()[name = tensor("op_3397"), val = tensor([1, 8, 128, 188])]; + tensor var_3398_cast_fp16 = reshape(shape = var_3397, x = key_17_cast_fp16)[name = tensor("op_3398_cast_fp16")]; + tensor mh_w_33_transpose_x_0 = const()[name = tensor("mh_w_33_transpose_x_0"), val = tensor(true)]; + tensor mh_w_33_transpose_y_0 = const()[name = tensor("mh_w_33_transpose_y_0"), val = tensor(false)]; + tensor mh_w_33_cast_fp16 = matmul(transpose_x = mh_w_33_transpose_x_0, transpose_y = mh_w_33_transpose_y_0, x = var_3394_cast_fp16, y = var_3398_cast_fp16)[name = tensor("mh_w_33_cast_fp16")]; + tensor mh_w_35_cast_fp16 = add(x = mh_w_33_cast_fp16, y = qk_mask_17_cast_fp16)[name = tensor("mh_w_35_cast_fp16")]; + tensor var_3402_cast_fp16 = softmax(axis = var_3189, x = mh_w_35_cast_fp16)[name = tensor("op_3402_cast_fp16")]; + tensor var_3403 = const()[name = tensor("op_3403"), val = tensor([1, 8, 128, 188])]; + tensor var_3404_cast_fp16 = reshape(shape = var_3403, x = value_17_cast_fp16)[name = tensor("op_3404_cast_fp16")]; + tensor attn_17_transpose_x_0 = const()[name = tensor("attn_17_transpose_x_0"), val = tensor(false)]; + tensor attn_17_transpose_y_0 = const()[name = tensor("attn_17_transpose_y_0"), val = tensor(true)]; + tensor attn_17_cast_fp16 = matmul(transpose_x = attn_17_transpose_x_0, transpose_y = attn_17_transpose_y_0, x = var_3404_cast_fp16, y = var_3402_cast_fp16)[name = tensor("attn_17_cast_fp16")]; + tensor var_3407 = const()[name = tensor("op_3407"), val = tensor([1, 1024, 1, 188])]; + tensor input_231_cast_fp16 = reshape(shape = var_3407, x = attn_17_cast_fp16)[name = tensor("input_231_cast_fp16")]; + tensor var_3417_pad_type_0 = const()[name = tensor("op_3417_pad_type_0"), val = tensor("valid")]; + tensor var_3417_strides_0 = const()[name = tensor("op_3417_strides_0"), val = tensor([1, 1])]; + tensor var_3417_pad_0 = const()[name = tensor("op_3417_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3417_dilations_0 = const()[name = tensor("op_3417_dilations_0"), val = tensor([1, 1])]; + tensor var_3417_groups_0 = const()[name = tensor("op_3417_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(200703360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(201489856))), name = tensor("layers_8_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3417_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3417_dilations_0, groups = var_3417_groups_0, pad = var_3417_pad_0, pad_type = var_3417_pad_type_0, strides = var_3417_strides_0, weight = layers_8_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_231_cast_fp16)[name = tensor("op_3417_cast_fp16")]; + tensor var_3423_pad_type_0 = const()[name = tensor("op_3423_pad_type_0"), val = tensor("valid")]; + tensor var_3423_strides_0 = const()[name = tensor("op_3423_strides_0"), val = tensor([1, 1])]; + tensor var_3423_pad_0 = const()[name = tensor("op_3423_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3423_dilations_0 = const()[name = tensor("op_3423_dilations_0"), val = tensor([1, 1])]; + tensor var_3423_groups_0 = const()[name = tensor("op_3423_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(201521856))), name = tensor("layers_8_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(201490048))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3423_cast_fp16 = conv(dilations = var_3423_dilations_0, groups = var_3423_groups_0, pad = var_3423_pad_0, pad_type = var_3423_pad_type_0, strides = var_3423_strides_0, weight = layers_8_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_231_cast_fp16)[name = tensor("op_3423_cast_fp16")]; + tensor obj_37_cast_fp16 = add(x = var_3417_cast_fp16, y = var_3423_cast_fp16)[name = tensor("obj_37_cast_fp16")]; + tensor inputs_85_cast_fp16 = add(x = inputs_83_cast_fp16, y = obj_37_cast_fp16)[name = tensor("inputs_85_cast_fp16")]; + tensor out_85_axes_0 = const()[name = tensor("out_85_axes_0"), val = tensor([1])]; + tensor var_3434_to_fp16 = const()[name = tensor("op_3434_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_85_cast_fp16 = layer_norm(axes = out_85_axes_0, epsilon = var_3434_to_fp16, x = inputs_85_cast_fp16)[name = tensor("out_85_cast_fp16")]; + tensor input_233_gamma_0_to_fp16 = const()[name = tensor("input_233_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(201652992)))]; + tensor input_233_beta_0_to_fp16 = const()[name = tensor("input_233_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(201655104)))]; + tensor input_233_epsilon_0_to_fp16 = const()[name = tensor("input_233_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_233_cast_fp16 = batch_norm(beta = input_233_beta_0_to_fp16, epsilon = input_233_epsilon_0_to_fp16, gamma = input_233_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_85_cast_fp16)[name = tensor("input_233_cast_fp16")]; + tensor var_3455_pad_type_0 = const()[name = tensor("op_3455_pad_type_0"), val = tensor("valid")]; + tensor var_3455_strides_0 = const()[name = tensor("op_3455_strides_0"), val = tensor([1, 1])]; + tensor var_3455_pad_0 = const()[name = tensor("op_3455_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3455_dilations_0 = const()[name = tensor("op_3455_dilations_0"), val = tensor([1, 1])]; + tensor var_3455_groups_0 = const()[name = tensor("op_3455_groups_0"), val = tensor(1)]; + tensor layers_8_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(201657216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203230144))), name = tensor("layers_8_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_3455_cast_fp16 = conv(dilations = var_3455_dilations_0, groups = var_3455_groups_0, pad = var_3455_pad_0, pad_type = var_3455_pad_type_0, strides = var_3455_strides_0, weight = layers_8_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_233_cast_fp16)[name = tensor("op_3455_cast_fp16")]; + tensor var_3461_pad_type_0 = const()[name = tensor("op_3461_pad_type_0"), val = tensor("valid")]; + tensor var_3461_strides_0 = const()[name = tensor("op_3461_strides_0"), val = tensor([1, 1])]; + tensor var_3461_pad_0 = const()[name = tensor("op_3461_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3461_dilations_0 = const()[name = tensor("op_3461_dilations_0"), val = tensor([1, 1])]; + tensor var_3461_groups_0 = const()[name = tensor("op_3461_groups_0"), val = tensor(1)]; + tensor layers_8_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203290880))), name = tensor("layers_8_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203230336))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_3461_cast_fp16 = conv(dilations = var_3461_dilations_0, groups = var_3461_groups_0, pad = var_3461_pad_0, pad_type = var_3461_pad_type_0, strides = var_3461_strides_0, weight = layers_8_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_233_cast_fp16)[name = tensor("op_3461_cast_fp16")]; + tensor input_235_cast_fp16 = add(x = var_3455_cast_fp16, y = var_3461_cast_fp16)[name = tensor("input_235_cast_fp16")]; + tensor input_237_split_num_splits_0 = const()[name = tensor("input_237_split_num_splits_0"), val = tensor(2)]; + tensor input_237_split_axis_0 = const()[name = tensor("input_237_split_axis_0"), val = tensor(1)]; + tensor input_237_split_cast_fp16_0, tensor input_237_split_cast_fp16_1 = split(axis = input_237_split_axis_0, num_splits = input_237_split_num_splits_0, x = input_235_cast_fp16)[name = tensor("input_237_split_cast_fp16")]; + tensor input_237_split_1_sigmoid_cast_fp16 = sigmoid(x = input_237_split_cast_fp16_1)[name = tensor("input_237_split_1_sigmoid_cast_fp16")]; + tensor input_237_cast_fp16 = mul(x = input_237_split_cast_fp16_0, y = input_237_split_1_sigmoid_cast_fp16)[name = tensor("input_237_cast_fp16")]; + tensor input_239_pad_type_0 = const()[name = tensor("input_239_pad_type_0"), val = tensor("custom")]; + tensor input_239_pad_0 = const()[name = tensor("input_239_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_239_groups_0 = const()[name = tensor("input_239_groups_0"), val = tensor(1024)]; + tensor input_239_strides_0 = const()[name = tensor("input_239_strides_0"), val = tensor([1, 1])]; + tensor input_239_dilations_0 = const()[name = tensor("input_239_dilations_0"), val = tensor([1, 1])]; + tensor const_284_to_fp16 = const()[name = tensor("const_284_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203553088)))]; + tensor const_285_to_fp16 = const()[name = tensor("const_285_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203571584)))]; + tensor input_241_cast_fp16 = conv(bias = const_285_to_fp16, dilations = input_239_dilations_0, groups = input_239_groups_0, pad = input_239_pad_0, pad_type = input_239_pad_type_0, strides = input_239_strides_0, weight = const_284_to_fp16, x = input_237_cast_fp16)[name = tensor("input_241_cast_fp16")]; + tensor input_243_cast_fp16 = silu(x = input_241_cast_fp16)[name = tensor("input_243_cast_fp16")]; + tensor var_3483_pad_type_0 = const()[name = tensor("op_3483_pad_type_0"), val = tensor("valid")]; + tensor var_3483_strides_0 = const()[name = tensor("op_3483_strides_0"), val = tensor([1, 1])]; + tensor var_3483_pad_0 = const()[name = tensor("op_3483_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3483_dilations_0 = const()[name = tensor("op_3483_dilations_0"), val = tensor([1, 1])]; + tensor var_3483_groups_0 = const()[name = tensor("op_3483_groups_0"), val = tensor(1)]; + tensor layers_8_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203573696))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204360192))), name = tensor("layers_8_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3483_cast_fp16 = conv(dilations = var_3483_dilations_0, groups = var_3483_groups_0, pad = var_3483_pad_0, pad_type = var_3483_pad_type_0, strides = var_3483_strides_0, weight = layers_8_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_243_cast_fp16)[name = tensor("op_3483_cast_fp16")]; + tensor var_3489_pad_type_0 = const()[name = tensor("op_3489_pad_type_0"), val = tensor("valid")]; + tensor var_3489_strides_0 = const()[name = tensor("op_3489_strides_0"), val = tensor([1, 1])]; + tensor var_3489_pad_0 = const()[name = tensor("op_3489_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3489_dilations_0 = const()[name = tensor("op_3489_dilations_0"), val = tensor([1, 1])]; + tensor var_3489_groups_0 = const()[name = tensor("op_3489_groups_0"), val = tensor(1)]; + tensor layers_8_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204390464))), name = tensor("layers_8_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204360384))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3489_cast_fp16 = conv(dilations = var_3489_dilations_0, groups = var_3489_groups_0, pad = var_3489_pad_0, pad_type = var_3489_pad_type_0, strides = var_3489_strides_0, weight = layers_8_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_243_cast_fp16)[name = tensor("op_3489_cast_fp16")]; + tensor x_53_cast_fp16 = add(x = var_3483_cast_fp16, y = var_3489_cast_fp16)[name = tensor("x_53_cast_fp16")]; + tensor inputs_87_cast_fp16 = add(x = inputs_85_cast_fp16, y = x_53_cast_fp16)[name = tensor("inputs_87_cast_fp16")]; + tensor out_87_axes_0 = const()[name = tensor("out_87_axes_0"), val = tensor([1])]; + tensor var_3500_to_fp16 = const()[name = tensor("op_3500_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_87_cast_fp16 = layer_norm(axes = out_87_axes_0, epsilon = var_3500_to_fp16, x = inputs_87_cast_fp16)[name = tensor("out_87_cast_fp16")]; + tensor input_245_gamma_0_to_fp16 = const()[name = tensor("input_245_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204521600)))]; + tensor input_245_beta_0_to_fp16 = const()[name = tensor("input_245_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204523712)))]; + tensor input_245_epsilon_0_to_fp16 = const()[name = tensor("input_245_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_245_cast_fp16 = batch_norm(beta = input_245_beta_0_to_fp16, epsilon = input_245_epsilon_0_to_fp16, gamma = input_245_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_87_cast_fp16)[name = tensor("input_245_cast_fp16")]; + tensor var_3520_pad_type_0 = const()[name = tensor("op_3520_pad_type_0"), val = tensor("valid")]; + tensor var_3520_strides_0 = const()[name = tensor("op_3520_strides_0"), val = tensor([1, 1])]; + tensor var_3520_pad_0 = const()[name = tensor("op_3520_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3520_dilations_0 = const()[name = tensor("op_3520_dilations_0"), val = tensor([1, 1])]; + tensor var_3520_groups_0 = const()[name = tensor("op_3520_groups_0"), val = tensor(1)]; + tensor layers_8_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(204525824))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(207671616))), name = tensor("layers_8_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_3520_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_3520_dilations_0, groups = var_3520_groups_0, pad = var_3520_pad_0, pad_type = var_3520_pad_type_0, strides = var_3520_strides_0, weight = layers_8_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_245_cast_fp16)[name = tensor("op_3520_cast_fp16")]; + tensor var_3526_pad_type_0 = const()[name = tensor("op_3526_pad_type_0"), val = tensor("valid")]; + tensor var_3526_strides_0 = const()[name = tensor("op_3526_strides_0"), val = tensor([1, 1])]; + tensor var_3526_pad_0 = const()[name = tensor("op_3526_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3526_dilations_0 = const()[name = tensor("op_3526_dilations_0"), val = tensor([1, 1])]; + tensor var_3526_groups_0 = const()[name = tensor("op_3526_groups_0"), val = tensor(1)]; + tensor layers_8_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(207820096))), name = tensor("layers_8_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(207671808))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_3526_cast_fp16 = conv(dilations = var_3526_dilations_0, groups = var_3526_groups_0, pad = var_3526_pad_0, pad_type = var_3526_pad_type_0, strides = var_3526_strides_0, weight = layers_8_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_245_cast_fp16)[name = tensor("op_3526_cast_fp16")]; + tensor input_247_cast_fp16 = add(x = var_3520_cast_fp16, y = var_3526_cast_fp16)[name = tensor("input_247_cast_fp16")]; + tensor input_249_cast_fp16 = silu(x = input_247_cast_fp16)[name = tensor("input_249_cast_fp16")]; + tensor var_3537_pad_type_0 = const()[name = tensor("op_3537_pad_type_0"), val = tensor("valid")]; + tensor var_3537_strides_0 = const()[name = tensor("op_3537_strides_0"), val = tensor([1, 1])]; + tensor var_3537_pad_0 = const()[name = tensor("op_3537_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3537_dilations_0 = const()[name = tensor("op_3537_dilations_0"), val = tensor([1, 1])]; + tensor var_3537_groups_0 = const()[name = tensor("op_3537_groups_0"), val = tensor(1)]; + tensor layers_8_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(208344448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211490240))), name = tensor("layers_8_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_3537_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3537_dilations_0, groups = var_3537_groups_0, pad = var_3537_pad_0, pad_type = var_3537_pad_type_0, strides = var_3537_strides_0, weight = layers_8_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_249_cast_fp16)[name = tensor("op_3537_cast_fp16")]; + tensor var_3543_pad_type_0 = const()[name = tensor("op_3543_pad_type_0"), val = tensor("valid")]; + tensor var_3543_strides_0 = const()[name = tensor("op_3543_strides_0"), val = tensor([1, 1])]; + tensor var_3543_pad_0 = const()[name = tensor("op_3543_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3543_dilations_0 = const()[name = tensor("op_3543_dilations_0"), val = tensor([1, 1])]; + tensor var_3543_groups_0 = const()[name = tensor("op_3543_groups_0"), val = tensor(1)]; + tensor layers_8_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211632896))), name = tensor("layers_8_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211490432))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_3543_cast_fp16 = conv(dilations = var_3543_dilations_0, groups = var_3543_groups_0, pad = var_3543_pad_0, pad_type = var_3543_pad_type_0, strides = var_3543_strides_0, weight = layers_8_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_249_cast_fp16)[name = tensor("op_3543_cast_fp16")]; + tensor x_55_cast_fp16 = add(x = var_3537_cast_fp16, y = var_3543_cast_fp16)[name = tensor("x_55_cast_fp16")]; + tensor var_3545_to_fp16 = const()[name = tensor("op_3545_to_fp16"), val = tensor(0x1p-1)]; + tensor var_3546_cast_fp16 = mul(x = x_55_cast_fp16, y = var_3545_to_fp16)[name = tensor("op_3546_cast_fp16")]; + tensor inputs_89_cast_fp16 = add(x = inputs_87_cast_fp16, y = var_3546_cast_fp16)[name = tensor("inputs_89_cast_fp16")]; + tensor out_89_axes_0 = const()[name = tensor("out_89_axes_0"), val = tensor([1])]; + tensor var_3556_to_fp16 = const()[name = tensor("op_3556_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_89_cast_fp16 = layer_norm(axes = out_89_axes_0, epsilon = var_3556_to_fp16, x = inputs_89_cast_fp16)[name = tensor("out_89_cast_fp16")]; + tensor inputs_91_gamma_0_to_fp16 = const()[name = tensor("inputs_91_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(212157248)))]; + tensor inputs_91_beta_0_to_fp16 = const()[name = tensor("inputs_91_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(212159360)))]; + tensor inputs_91_epsilon_0_to_fp16 = const()[name = tensor("inputs_91_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_91_cast_fp16 = batch_norm(beta = inputs_91_beta_0_to_fp16, epsilon = inputs_91_epsilon_0_to_fp16, gamma = inputs_91_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_89_cast_fp16)[name = tensor("inputs_91_cast_fp16")]; + tensor var_3570 = const()[name = tensor("op_3570"), val = tensor(3)]; + tensor out_91_axes_0 = const()[name = tensor("out_91_axes_0"), val = tensor([1])]; + tensor var_3601_to_fp16 = const()[name = tensor("op_3601_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_91_cast_fp16 = layer_norm(axes = out_91_axes_0, epsilon = var_3601_to_fp16, x = inputs_91_cast_fp16)[name = tensor("out_91_cast_fp16")]; + tensor input_251_gamma_0_to_fp16 = const()[name = tensor("input_251_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(212161472)))]; + tensor input_251_beta_0_to_fp16 = const()[name = tensor("input_251_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(212163584)))]; + tensor input_251_epsilon_0_to_fp16 = const()[name = tensor("input_251_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_251_cast_fp16 = batch_norm(beta = input_251_beta_0_to_fp16, epsilon = input_251_epsilon_0_to_fp16, gamma = input_251_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_91_cast_fp16)[name = tensor("input_251_cast_fp16")]; + tensor var_3621_pad_type_0 = const()[name = tensor("op_3621_pad_type_0"), val = tensor("valid")]; + tensor var_3621_strides_0 = const()[name = tensor("op_3621_strides_0"), val = tensor([1, 1])]; + tensor var_3621_pad_0 = const()[name = tensor("op_3621_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3621_dilations_0 = const()[name = tensor("op_3621_dilations_0"), val = tensor([1, 1])]; + tensor var_3621_groups_0 = const()[name = tensor("op_3621_groups_0"), val = tensor(1)]; + tensor layers_9_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(212165696))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215311488))), name = tensor("layers_9_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_3621_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_3621_dilations_0, groups = var_3621_groups_0, pad = var_3621_pad_0, pad_type = var_3621_pad_type_0, strides = var_3621_strides_0, weight = layers_9_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_251_cast_fp16)[name = tensor("op_3621_cast_fp16")]; + tensor var_3627_pad_type_0 = const()[name = tensor("op_3627_pad_type_0"), val = tensor("valid")]; + tensor var_3627_strides_0 = const()[name = tensor("op_3627_strides_0"), val = tensor([1, 1])]; + tensor var_3627_pad_0 = const()[name = tensor("op_3627_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3627_dilations_0 = const()[name = tensor("op_3627_dilations_0"), val = tensor([1, 1])]; + tensor var_3627_groups_0 = const()[name = tensor("op_3627_groups_0"), val = tensor(1)]; + tensor layers_9_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215469504))), name = tensor("layers_9_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215311680))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_3627_cast_fp16 = conv(dilations = var_3627_dilations_0, groups = var_3627_groups_0, pad = var_3627_pad_0, pad_type = var_3627_pad_type_0, strides = var_3627_strides_0, weight = layers_9_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_251_cast_fp16)[name = tensor("op_3627_cast_fp16")]; + tensor input_253_cast_fp16 = add(x = var_3621_cast_fp16, y = var_3627_cast_fp16)[name = tensor("input_253_cast_fp16")]; + tensor input_255_cast_fp16 = silu(x = input_253_cast_fp16)[name = tensor("input_255_cast_fp16")]; + tensor var_3638_pad_type_0 = const()[name = tensor("op_3638_pad_type_0"), val = tensor("valid")]; + tensor var_3638_strides_0 = const()[name = tensor("op_3638_strides_0"), val = tensor([1, 1])]; + tensor var_3638_pad_0 = const()[name = tensor("op_3638_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3638_dilations_0 = const()[name = tensor("op_3638_dilations_0"), val = tensor([1, 1])]; + tensor var_3638_groups_0 = const()[name = tensor("op_3638_groups_0"), val = tensor(1)]; + tensor layers_9_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215993856))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219139648))), name = tensor("layers_9_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_3638_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3638_dilations_0, groups = var_3638_groups_0, pad = var_3638_pad_0, pad_type = var_3638_pad_type_0, strides = var_3638_strides_0, weight = layers_9_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_255_cast_fp16)[name = tensor("op_3638_cast_fp16")]; + tensor var_3644_pad_type_0 = const()[name = tensor("op_3644_pad_type_0"), val = tensor("valid")]; + tensor var_3644_strides_0 = const()[name = tensor("op_3644_strides_0"), val = tensor([1, 1])]; + tensor var_3644_pad_0 = const()[name = tensor("op_3644_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3644_dilations_0 = const()[name = tensor("op_3644_dilations_0"), val = tensor([1, 1])]; + tensor var_3644_groups_0 = const()[name = tensor("op_3644_groups_0"), val = tensor(1)]; + tensor layers_9_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219301568))), name = tensor("layers_9_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219139840))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_3644_cast_fp16 = conv(dilations = var_3644_dilations_0, groups = var_3644_groups_0, pad = var_3644_pad_0, pad_type = var_3644_pad_type_0, strides = var_3644_strides_0, weight = layers_9_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_255_cast_fp16)[name = tensor("op_3644_cast_fp16")]; + tensor x_57_cast_fp16 = add(x = var_3638_cast_fp16, y = var_3644_cast_fp16)[name = tensor("x_57_cast_fp16")]; + tensor var_3646_to_fp16 = const()[name = tensor("op_3646_to_fp16"), val = tensor(0x1p-1)]; + tensor var_3647_cast_fp16 = mul(x = x_57_cast_fp16, y = var_3646_to_fp16)[name = tensor("op_3647_cast_fp16")]; + tensor inputs_93_cast_fp16 = add(x = inputs_91_cast_fp16, y = var_3647_cast_fp16)[name = tensor("inputs_93_cast_fp16")]; + tensor out_93_axes_0 = const()[name = tensor("out_93_axes_0"), val = tensor([1])]; + tensor var_3657_to_fp16 = const()[name = tensor("op_3657_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_93_cast_fp16 = layer_norm(axes = out_93_axes_0, epsilon = var_3657_to_fp16, x = inputs_93_cast_fp16)[name = tensor("out_93_cast_fp16")]; + tensor obj_39_gamma_0_to_fp16 = const()[name = tensor("obj_39_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219825920)))]; + tensor obj_39_beta_0_to_fp16 = const()[name = tensor("obj_39_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219828032)))]; + tensor obj_39_epsilon_0_to_fp16 = const()[name = tensor("obj_39_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_39_cast_fp16 = batch_norm(beta = obj_39_beta_0_to_fp16, epsilon = obj_39_epsilon_0_to_fp16, gamma = obj_39_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_93_cast_fp16)[name = tensor("obj_39_cast_fp16")]; + tensor var_3682_pad_type_0 = const()[name = tensor("op_3682_pad_type_0"), val = tensor("valid")]; + tensor var_3682_strides_0 = const()[name = tensor("op_3682_strides_0"), val = tensor([1, 1])]; + tensor var_3682_pad_0 = const()[name = tensor("op_3682_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3682_dilations_0 = const()[name = tensor("op_3682_dilations_0"), val = tensor([1, 1])]; + tensor var_3682_groups_0 = const()[name = tensor("op_3682_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219830144))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220616640))), name = tensor("layers_9_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3682_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3682_dilations_0, groups = var_3682_groups_0, pad = var_3682_pad_0, pad_type = var_3682_pad_type_0, strides = var_3682_strides_0, weight = layers_9_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_39_cast_fp16)[name = tensor("op_3682_cast_fp16")]; + tensor var_3688_pad_type_0 = const()[name = tensor("op_3688_pad_type_0"), val = tensor("valid")]; + tensor var_3688_strides_0 = const()[name = tensor("op_3688_strides_0"), val = tensor([1, 1])]; + tensor var_3688_pad_0 = const()[name = tensor("op_3688_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3688_dilations_0 = const()[name = tensor("op_3688_dilations_0"), val = tensor([1, 1])]; + tensor var_3688_groups_0 = const()[name = tensor("op_3688_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220656384))), name = tensor("layers_9_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220616832))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3688_cast_fp16 = conv(dilations = var_3688_dilations_0, groups = var_3688_groups_0, pad = var_3688_pad_0, pad_type = var_3688_pad_type_0, strides = var_3688_strides_0, weight = layers_9_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_39_cast_fp16)[name = tensor("op_3688_cast_fp16")]; + tensor query_37_cast_fp16 = add(x = var_3682_cast_fp16, y = var_3688_cast_fp16)[name = tensor("query_37_cast_fp16")]; + tensor var_3697_pad_type_0 = const()[name = tensor("op_3697_pad_type_0"), val = tensor("valid")]; + tensor var_3697_strides_0 = const()[name = tensor("op_3697_strides_0"), val = tensor([1, 1])]; + tensor var_3697_pad_0 = const()[name = tensor("op_3697_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3697_dilations_0 = const()[name = tensor("op_3697_dilations_0"), val = tensor([1, 1])]; + tensor var_3697_groups_0 = const()[name = tensor("op_3697_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220787520))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(221574016))), name = tensor("layers_9_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3697_cast_fp16 = conv(dilations = var_3697_dilations_0, groups = var_3697_groups_0, pad = var_3697_pad_0, pad_type = var_3697_pad_type_0, strides = var_3697_strides_0, weight = layers_9_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_39_cast_fp16)[name = tensor("op_3697_cast_fp16")]; + tensor var_3703_pad_type_0 = const()[name = tensor("op_3703_pad_type_0"), val = tensor("valid")]; + tensor var_3703_strides_0 = const()[name = tensor("op_3703_strides_0"), val = tensor([1, 1])]; + tensor var_3703_pad_0 = const()[name = tensor("op_3703_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3703_dilations_0 = const()[name = tensor("op_3703_dilations_0"), val = tensor([1, 1])]; + tensor var_3703_groups_0 = const()[name = tensor("op_3703_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(221632512))), name = tensor("layers_9_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(221574208))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3703_cast_fp16 = conv(dilations = var_3703_dilations_0, groups = var_3703_groups_0, pad = var_3703_pad_0, pad_type = var_3703_pad_type_0, strides = var_3703_strides_0, weight = layers_9_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_39_cast_fp16)[name = tensor("op_3703_cast_fp16")]; + tensor key_19_cast_fp16 = add(x = var_3697_cast_fp16, y = var_3703_cast_fp16)[name = tensor("key_19_cast_fp16")]; + tensor var_3713_pad_type_0 = const()[name = tensor("op_3713_pad_type_0"), val = tensor("valid")]; + tensor var_3713_strides_0 = const()[name = tensor("op_3713_strides_0"), val = tensor([1, 1])]; + tensor var_3713_pad_0 = const()[name = tensor("op_3713_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3713_dilations_0 = const()[name = tensor("op_3713_dilations_0"), val = tensor([1, 1])]; + tensor var_3713_groups_0 = const()[name = tensor("op_3713_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(221763648))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222550144))), name = tensor("layers_9_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3713_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3713_dilations_0, groups = var_3713_groups_0, pad = var_3713_pad_0, pad_type = var_3713_pad_type_0, strides = var_3713_strides_0, weight = layers_9_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_39_cast_fp16)[name = tensor("op_3713_cast_fp16")]; + tensor var_3719_pad_type_0 = const()[name = tensor("op_3719_pad_type_0"), val = tensor("valid")]; + tensor var_3719_strides_0 = const()[name = tensor("op_3719_strides_0"), val = tensor([1, 1])]; + tensor var_3719_pad_0 = const()[name = tensor("op_3719_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3719_dilations_0 = const()[name = tensor("op_3719_dilations_0"), val = tensor([1, 1])]; + tensor var_3719_groups_0 = const()[name = tensor("op_3719_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222581568))), name = tensor("layers_9_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222550336))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3719_cast_fp16 = conv(dilations = var_3719_dilations_0, groups = var_3719_groups_0, pad = var_3719_pad_0, pad_type = var_3719_pad_type_0, strides = var_3719_strides_0, weight = layers_9_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_39_cast_fp16)[name = tensor("op_3719_cast_fp16")]; + tensor value_19_cast_fp16 = add(x = var_3713_cast_fp16, y = var_3719_cast_fp16)[name = tensor("value_19_cast_fp16")]; + tensor var_3722_to_fp16 = const()[name = tensor("op_3722_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222712704)))]; + tensor query_39_cast_fp16 = add(x = query_37_cast_fp16, y = var_3722_to_fp16)[name = tensor("query_39_cast_fp16")]; + tensor var_3725_to_fp16 = const()[name = tensor("op_3725_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222714816)))]; + tensor q_with_bias_v_19_cast_fp16 = add(x = query_37_cast_fp16, y = var_3725_to_fp16)[name = tensor("q_with_bias_v_19_cast_fp16")]; + tensor var_3735_pad_type_0 = const()[name = tensor("op_3735_pad_type_0"), val = tensor("valid")]; + tensor var_3735_strides_0 = const()[name = tensor("op_3735_strides_0"), val = tensor([1, 1])]; + tensor var_3735_pad_0 = const()[name = tensor("op_3735_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3735_dilations_0 = const()[name = tensor("op_3735_dilations_0"), val = tensor([1, 1])]; + tensor var_3735_groups_0 = const()[name = tensor("op_3735_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(222716928))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(223503424))), name = tensor("layers_9_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3735_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3735_dilations_0, groups = var_3735_groups_0, pad = var_3735_pad_0, pad_type = var_3735_pad_type_0, strides = var_3735_strides_0, weight = layers_9_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_3735_cast_fp16")]; + tensor var_3741_pad_type_0 = const()[name = tensor("op_3741_pad_type_0"), val = tensor("valid")]; + tensor var_3741_strides_0 = const()[name = tensor("op_3741_strides_0"), val = tensor([1, 1])]; + tensor var_3741_pad_0 = const()[name = tensor("op_3741_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3741_dilations_0 = const()[name = tensor("op_3741_dilations_0"), val = tensor([1, 1])]; + tensor var_3741_groups_0 = const()[name = tensor("op_3741_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(223567040))), name = tensor("layers_9_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(223503616))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3741_cast_fp16 = conv(dilations = var_3741_dilations_0, groups = var_3741_groups_0, pad = var_3741_pad_0, pad_type = var_3741_pad_type_0, strides = var_3741_strides_0, weight = layers_9_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_3741_cast_fp16")]; + tensor p_19_cast_fp16 = add(x = var_3735_cast_fp16, y = var_3741_cast_fp16)[name = tensor("p_19_cast_fp16")]; + tensor var_3745 = const()[name = tensor("op_3745"), val = tensor([1, 8, 128, 188])]; + tensor var_3746_cast_fp16 = reshape(shape = var_3745, x = q_with_bias_v_19_cast_fp16)[name = tensor("op_3746_cast_fp16")]; + tensor var_3747 = const()[name = tensor("op_3747"), val = tensor([1, 8, 128, -1])]; + tensor var_3748_cast_fp16 = reshape(shape = var_3747, x = p_19_cast_fp16)[name = tensor("op_3748_cast_fp16")]; + tensor matrix_bd_73_transpose_x_0 = const()[name = tensor("matrix_bd_73_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_73_transpose_y_0 = const()[name = tensor("matrix_bd_73_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_73_cast_fp16 = matmul(transpose_x = matrix_bd_73_transpose_x_0, transpose_y = matrix_bd_73_transpose_y_0, x = var_3746_cast_fp16, y = var_3748_cast_fp16)[name = tensor("matrix_bd_73_cast_fp16")]; + tensor matrix_bd_75_pad_0 = const()[name = tensor("matrix_bd_75_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_75_mode_0 = const()[name = tensor("matrix_bd_75_mode_0"), val = tensor("constant")]; + tensor const_109_to_fp16 = const()[name = tensor("const_109_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_75_cast_fp16 = pad(constant_val = const_109_to_fp16, mode = matrix_bd_75_mode_0, pad = matrix_bd_75_pad_0, x = matrix_bd_73_cast_fp16)[name = tensor("matrix_bd_75_cast_fp16")]; + tensor var_3757 = const()[name = tensor("op_3757"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_77_cast_fp16 = reshape(shape = var_3757, x = matrix_bd_75_cast_fp16)[name = tensor("matrix_bd_77_cast_fp16")]; + tensor var_3761_begin_0 = const()[name = tensor("op_3761_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_3761_end_0 = const()[name = tensor("op_3761_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_3761_end_mask_0 = const()[name = tensor("op_3761_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3761_cast_fp16 = slice_by_index(begin = var_3761_begin_0, end = var_3761_end_0, end_mask = var_3761_end_mask_0, x = matrix_bd_77_cast_fp16)[name = tensor("op_3761_cast_fp16")]; + tensor var_3762 = const()[name = tensor("op_3762"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_79_cast_fp16 = reshape(shape = var_3762, x = var_3761_cast_fp16)[name = tensor("matrix_bd_79_cast_fp16")]; + tensor var_3767_begin_0 = const()[name = tensor("op_3767_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3767_end_0 = const()[name = tensor("op_3767_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_3767_end_mask_0 = const()[name = tensor("op_3767_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3767_cast_fp16 = slice_by_index(begin = var_3767_begin_0, end = var_3767_end_0, end_mask = var_3767_end_mask_0, x = matrix_bd_79_cast_fp16)[name = tensor("op_3767_cast_fp16")]; + tensor var_3768_to_fp16 = const()[name = tensor("op_3768_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_19_cast_fp16 = mul(x = var_3767_cast_fp16, y = var_3768_to_fp16)[name = tensor("qk_mask_19_cast_fp16")]; + tensor var_3772 = const()[name = tensor("op_3772"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_19_cast_fp16 = reshape(shape = var_3772, x = query_39_cast_fp16)[name = tensor("mh_q_19_cast_fp16")]; + tensor var_3774_to_fp16 = const()[name = tensor("op_3774_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_3775_cast_fp16 = mul(x = mh_q_19_cast_fp16, y = var_3774_to_fp16)[name = tensor("op_3775_cast_fp16")]; + tensor var_3778 = const()[name = tensor("op_3778"), val = tensor([1, 8, 128, 188])]; + tensor var_3779_cast_fp16 = reshape(shape = var_3778, x = key_19_cast_fp16)[name = tensor("op_3779_cast_fp16")]; + tensor mh_w_37_transpose_x_0 = const()[name = tensor("mh_w_37_transpose_x_0"), val = tensor(true)]; + tensor mh_w_37_transpose_y_0 = const()[name = tensor("mh_w_37_transpose_y_0"), val = tensor(false)]; + tensor mh_w_37_cast_fp16 = matmul(transpose_x = mh_w_37_transpose_x_0, transpose_y = mh_w_37_transpose_y_0, x = var_3775_cast_fp16, y = var_3779_cast_fp16)[name = tensor("mh_w_37_cast_fp16")]; + tensor mh_w_39_cast_fp16 = add(x = mh_w_37_cast_fp16, y = qk_mask_19_cast_fp16)[name = tensor("mh_w_39_cast_fp16")]; + tensor var_3783_cast_fp16 = softmax(axis = var_3570, x = mh_w_39_cast_fp16)[name = tensor("op_3783_cast_fp16")]; + tensor var_3784 = const()[name = tensor("op_3784"), val = tensor([1, 8, 128, 188])]; + tensor var_3785_cast_fp16 = reshape(shape = var_3784, x = value_19_cast_fp16)[name = tensor("op_3785_cast_fp16")]; + tensor attn_19_transpose_x_0 = const()[name = tensor("attn_19_transpose_x_0"), val = tensor(false)]; + tensor attn_19_transpose_y_0 = const()[name = tensor("attn_19_transpose_y_0"), val = tensor(true)]; + tensor attn_19_cast_fp16 = matmul(transpose_x = attn_19_transpose_x_0, transpose_y = attn_19_transpose_y_0, x = var_3785_cast_fp16, y = var_3783_cast_fp16)[name = tensor("attn_19_cast_fp16")]; + tensor var_3788 = const()[name = tensor("op_3788"), val = tensor([1, 1024, 1, 188])]; + tensor input_257_cast_fp16 = reshape(shape = var_3788, x = attn_19_cast_fp16)[name = tensor("input_257_cast_fp16")]; + tensor var_3798_pad_type_0 = const()[name = tensor("op_3798_pad_type_0"), val = tensor("valid")]; + tensor var_3798_strides_0 = const()[name = tensor("op_3798_strides_0"), val = tensor([1, 1])]; + tensor var_3798_pad_0 = const()[name = tensor("op_3798_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3798_dilations_0 = const()[name = tensor("op_3798_dilations_0"), val = tensor([1, 1])]; + tensor var_3798_groups_0 = const()[name = tensor("op_3798_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(223698176))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224484672))), name = tensor("layers_9_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3798_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3798_dilations_0, groups = var_3798_groups_0, pad = var_3798_pad_0, pad_type = var_3798_pad_type_0, strides = var_3798_strides_0, weight = layers_9_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_257_cast_fp16)[name = tensor("op_3798_cast_fp16")]; + tensor var_3804_pad_type_0 = const()[name = tensor("op_3804_pad_type_0"), val = tensor("valid")]; + tensor var_3804_strides_0 = const()[name = tensor("op_3804_strides_0"), val = tensor([1, 1])]; + tensor var_3804_pad_0 = const()[name = tensor("op_3804_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3804_dilations_0 = const()[name = tensor("op_3804_dilations_0"), val = tensor([1, 1])]; + tensor var_3804_groups_0 = const()[name = tensor("op_3804_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224515008))), name = tensor("layers_9_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224484864))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3804_cast_fp16 = conv(dilations = var_3804_dilations_0, groups = var_3804_groups_0, pad = var_3804_pad_0, pad_type = var_3804_pad_type_0, strides = var_3804_strides_0, weight = layers_9_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_257_cast_fp16)[name = tensor("op_3804_cast_fp16")]; + tensor obj_41_cast_fp16 = add(x = var_3798_cast_fp16, y = var_3804_cast_fp16)[name = tensor("obj_41_cast_fp16")]; + tensor inputs_95_cast_fp16 = add(x = inputs_93_cast_fp16, y = obj_41_cast_fp16)[name = tensor("inputs_95_cast_fp16")]; + tensor out_95_axes_0 = const()[name = tensor("out_95_axes_0"), val = tensor([1])]; + tensor var_3815_to_fp16 = const()[name = tensor("op_3815_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_95_cast_fp16 = layer_norm(axes = out_95_axes_0, epsilon = var_3815_to_fp16, x = inputs_95_cast_fp16)[name = tensor("out_95_cast_fp16")]; + tensor input_259_gamma_0_to_fp16 = const()[name = tensor("input_259_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224646144)))]; + tensor input_259_beta_0_to_fp16 = const()[name = tensor("input_259_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224648256)))]; + tensor input_259_epsilon_0_to_fp16 = const()[name = tensor("input_259_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_259_cast_fp16 = batch_norm(beta = input_259_beta_0_to_fp16, epsilon = input_259_epsilon_0_to_fp16, gamma = input_259_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_95_cast_fp16)[name = tensor("input_259_cast_fp16")]; + tensor var_3836_pad_type_0 = const()[name = tensor("op_3836_pad_type_0"), val = tensor("valid")]; + tensor var_3836_strides_0 = const()[name = tensor("op_3836_strides_0"), val = tensor([1, 1])]; + tensor var_3836_pad_0 = const()[name = tensor("op_3836_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3836_dilations_0 = const()[name = tensor("op_3836_dilations_0"), val = tensor([1, 1])]; + tensor var_3836_groups_0 = const()[name = tensor("op_3836_groups_0"), val = tensor(1)]; + tensor layers_9_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224650368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226223296))), name = tensor("layers_9_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_3836_cast_fp16 = conv(dilations = var_3836_dilations_0, groups = var_3836_groups_0, pad = var_3836_pad_0, pad_type = var_3836_pad_type_0, strides = var_3836_strides_0, weight = layers_9_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_259_cast_fp16)[name = tensor("op_3836_cast_fp16")]; + tensor var_3842_pad_type_0 = const()[name = tensor("op_3842_pad_type_0"), val = tensor("valid")]; + tensor var_3842_strides_0 = const()[name = tensor("op_3842_strides_0"), val = tensor([1, 1])]; + tensor var_3842_pad_0 = const()[name = tensor("op_3842_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3842_dilations_0 = const()[name = tensor("op_3842_dilations_0"), val = tensor([1, 1])]; + tensor var_3842_groups_0 = const()[name = tensor("op_3842_groups_0"), val = tensor(1)]; + tensor layers_9_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226283328))), name = tensor("layers_9_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226223488))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_3842_cast_fp16 = conv(dilations = var_3842_dilations_0, groups = var_3842_groups_0, pad = var_3842_pad_0, pad_type = var_3842_pad_type_0, strides = var_3842_strides_0, weight = layers_9_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_259_cast_fp16)[name = tensor("op_3842_cast_fp16")]; + tensor input_261_cast_fp16 = add(x = var_3836_cast_fp16, y = var_3842_cast_fp16)[name = tensor("input_261_cast_fp16")]; + tensor input_263_split_num_splits_0 = const()[name = tensor("input_263_split_num_splits_0"), val = tensor(2)]; + tensor input_263_split_axis_0 = const()[name = tensor("input_263_split_axis_0"), val = tensor(1)]; + tensor input_263_split_cast_fp16_0, tensor input_263_split_cast_fp16_1 = split(axis = input_263_split_axis_0, num_splits = input_263_split_num_splits_0, x = input_261_cast_fp16)[name = tensor("input_263_split_cast_fp16")]; + tensor input_263_split_1_sigmoid_cast_fp16 = sigmoid(x = input_263_split_cast_fp16_1)[name = tensor("input_263_split_1_sigmoid_cast_fp16")]; + tensor input_263_cast_fp16 = mul(x = input_263_split_cast_fp16_0, y = input_263_split_1_sigmoid_cast_fp16)[name = tensor("input_263_cast_fp16")]; + tensor input_265_pad_type_0 = const()[name = tensor("input_265_pad_type_0"), val = tensor("custom")]; + tensor input_265_pad_0 = const()[name = tensor("input_265_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_265_groups_0 = const()[name = tensor("input_265_groups_0"), val = tensor(1024)]; + tensor input_265_strides_0 = const()[name = tensor("input_265_strides_0"), val = tensor([1, 1])]; + tensor input_265_dilations_0 = const()[name = tensor("input_265_dilations_0"), val = tensor([1, 1])]; + tensor const_286_to_fp16 = const()[name = tensor("const_286_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226545536)))]; + tensor const_287_to_fp16 = const()[name = tensor("const_287_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226564032)))]; + tensor input_267_cast_fp16 = conv(bias = const_287_to_fp16, dilations = input_265_dilations_0, groups = input_265_groups_0, pad = input_265_pad_0, pad_type = input_265_pad_type_0, strides = input_265_strides_0, weight = const_286_to_fp16, x = input_263_cast_fp16)[name = tensor("input_267_cast_fp16")]; + tensor input_269_cast_fp16 = silu(x = input_267_cast_fp16)[name = tensor("input_269_cast_fp16")]; + tensor var_3864_pad_type_0 = const()[name = tensor("op_3864_pad_type_0"), val = tensor("valid")]; + tensor var_3864_strides_0 = const()[name = tensor("op_3864_strides_0"), val = tensor([1, 1])]; + tensor var_3864_pad_0 = const()[name = tensor("op_3864_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3864_dilations_0 = const()[name = tensor("op_3864_dilations_0"), val = tensor([1, 1])]; + tensor var_3864_groups_0 = const()[name = tensor("op_3864_groups_0"), val = tensor(1)]; + tensor layers_9_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(226566144))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227352640))), name = tensor("layers_9_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3864_cast_fp16 = conv(dilations = var_3864_dilations_0, groups = var_3864_groups_0, pad = var_3864_pad_0, pad_type = var_3864_pad_type_0, strides = var_3864_strides_0, weight = layers_9_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_269_cast_fp16)[name = tensor("op_3864_cast_fp16")]; + tensor var_3870_pad_type_0 = const()[name = tensor("op_3870_pad_type_0"), val = tensor("valid")]; + tensor var_3870_strides_0 = const()[name = tensor("op_3870_strides_0"), val = tensor([1, 1])]; + tensor var_3870_pad_0 = const()[name = tensor("op_3870_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3870_dilations_0 = const()[name = tensor("op_3870_dilations_0"), val = tensor([1, 1])]; + tensor var_3870_groups_0 = const()[name = tensor("op_3870_groups_0"), val = tensor(1)]; + tensor layers_9_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227384192))), name = tensor("layers_9_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227352832))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_3870_cast_fp16 = conv(dilations = var_3870_dilations_0, groups = var_3870_groups_0, pad = var_3870_pad_0, pad_type = var_3870_pad_type_0, strides = var_3870_strides_0, weight = layers_9_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_269_cast_fp16)[name = tensor("op_3870_cast_fp16")]; + tensor x_59_cast_fp16 = add(x = var_3864_cast_fp16, y = var_3870_cast_fp16)[name = tensor("x_59_cast_fp16")]; + tensor inputs_97_cast_fp16 = add(x = inputs_95_cast_fp16, y = x_59_cast_fp16)[name = tensor("inputs_97_cast_fp16")]; + tensor out_97_axes_0 = const()[name = tensor("out_97_axes_0"), val = tensor([1])]; + tensor var_3881_to_fp16 = const()[name = tensor("op_3881_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_97_cast_fp16 = layer_norm(axes = out_97_axes_0, epsilon = var_3881_to_fp16, x = inputs_97_cast_fp16)[name = tensor("out_97_cast_fp16")]; + tensor input_271_gamma_0_to_fp16 = const()[name = tensor("input_271_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227515328)))]; + tensor input_271_beta_0_to_fp16 = const()[name = tensor("input_271_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227517440)))]; + tensor input_271_epsilon_0_to_fp16 = const()[name = tensor("input_271_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_271_cast_fp16 = batch_norm(beta = input_271_beta_0_to_fp16, epsilon = input_271_epsilon_0_to_fp16, gamma = input_271_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_97_cast_fp16)[name = tensor("input_271_cast_fp16")]; + tensor var_3901_pad_type_0 = const()[name = tensor("op_3901_pad_type_0"), val = tensor("valid")]; + tensor var_3901_strides_0 = const()[name = tensor("op_3901_strides_0"), val = tensor([1, 1])]; + tensor var_3901_pad_0 = const()[name = tensor("op_3901_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3901_dilations_0 = const()[name = tensor("op_3901_dilations_0"), val = tensor([1, 1])]; + tensor var_3901_groups_0 = const()[name = tensor("op_3901_groups_0"), val = tensor(1)]; + tensor layers_9_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(227519552))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(230665344))), name = tensor("layers_9_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_3901_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_3901_dilations_0, groups = var_3901_groups_0, pad = var_3901_pad_0, pad_type = var_3901_pad_type_0, strides = var_3901_strides_0, weight = layers_9_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_271_cast_fp16)[name = tensor("op_3901_cast_fp16")]; + tensor var_3907_pad_type_0 = const()[name = tensor("op_3907_pad_type_0"), val = tensor("valid")]; + tensor var_3907_strides_0 = const()[name = tensor("op_3907_strides_0"), val = tensor([1, 1])]; + tensor var_3907_pad_0 = const()[name = tensor("op_3907_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3907_dilations_0 = const()[name = tensor("op_3907_dilations_0"), val = tensor([1, 1])]; + tensor var_3907_groups_0 = const()[name = tensor("op_3907_groups_0"), val = tensor(1)]; + tensor layers_9_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(230817024))), name = tensor("layers_9_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(230665536))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_3907_cast_fp16 = conv(dilations = var_3907_dilations_0, groups = var_3907_groups_0, pad = var_3907_pad_0, pad_type = var_3907_pad_type_0, strides = var_3907_strides_0, weight = layers_9_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_271_cast_fp16)[name = tensor("op_3907_cast_fp16")]; + tensor input_273_cast_fp16 = add(x = var_3901_cast_fp16, y = var_3907_cast_fp16)[name = tensor("input_273_cast_fp16")]; + tensor input_275_cast_fp16 = silu(x = input_273_cast_fp16)[name = tensor("input_275_cast_fp16")]; + tensor var_3918_pad_type_0 = const()[name = tensor("op_3918_pad_type_0"), val = tensor("valid")]; + tensor var_3918_strides_0 = const()[name = tensor("op_3918_strides_0"), val = tensor([1, 1])]; + tensor var_3918_pad_0 = const()[name = tensor("op_3918_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3918_dilations_0 = const()[name = tensor("op_3918_dilations_0"), val = tensor([1, 1])]; + tensor var_3918_groups_0 = const()[name = tensor("op_3918_groups_0"), val = tensor(1)]; + tensor layers_9_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(231341376))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234487168))), name = tensor("layers_9_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_3918_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3918_dilations_0, groups = var_3918_groups_0, pad = var_3918_pad_0, pad_type = var_3918_pad_type_0, strides = var_3918_strides_0, weight = layers_9_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_275_cast_fp16)[name = tensor("op_3918_cast_fp16")]; + tensor var_3924_pad_type_0 = const()[name = tensor("op_3924_pad_type_0"), val = tensor("valid")]; + tensor var_3924_strides_0 = const()[name = tensor("op_3924_strides_0"), val = tensor([1, 1])]; + tensor var_3924_pad_0 = const()[name = tensor("op_3924_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3924_dilations_0 = const()[name = tensor("op_3924_dilations_0"), val = tensor([1, 1])]; + tensor var_3924_groups_0 = const()[name = tensor("op_3924_groups_0"), val = tensor(1)]; + tensor layers_9_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234625088))), name = tensor("layers_9_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(234487360))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_3924_cast_fp16 = conv(dilations = var_3924_dilations_0, groups = var_3924_groups_0, pad = var_3924_pad_0, pad_type = var_3924_pad_type_0, strides = var_3924_strides_0, weight = layers_9_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_275_cast_fp16)[name = tensor("op_3924_cast_fp16")]; + tensor x_61_cast_fp16 = add(x = var_3918_cast_fp16, y = var_3924_cast_fp16)[name = tensor("x_61_cast_fp16")]; + tensor var_3926_to_fp16 = const()[name = tensor("op_3926_to_fp16"), val = tensor(0x1p-1)]; + tensor var_3927_cast_fp16 = mul(x = x_61_cast_fp16, y = var_3926_to_fp16)[name = tensor("op_3927_cast_fp16")]; + tensor inputs_99_cast_fp16 = add(x = inputs_97_cast_fp16, y = var_3927_cast_fp16)[name = tensor("inputs_99_cast_fp16")]; + tensor out_99_axes_0 = const()[name = tensor("out_99_axes_0"), val = tensor([1])]; + tensor var_3937_to_fp16 = const()[name = tensor("op_3937_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_99_cast_fp16 = layer_norm(axes = out_99_axes_0, epsilon = var_3937_to_fp16, x = inputs_99_cast_fp16)[name = tensor("out_99_cast_fp16")]; + tensor inputs_101_gamma_0_to_fp16 = const()[name = tensor("inputs_101_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235149440)))]; + tensor inputs_101_beta_0_to_fp16 = const()[name = tensor("inputs_101_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235151552)))]; + tensor inputs_101_epsilon_0_to_fp16 = const()[name = tensor("inputs_101_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_101_cast_fp16 = batch_norm(beta = inputs_101_beta_0_to_fp16, epsilon = inputs_101_epsilon_0_to_fp16, gamma = inputs_101_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_99_cast_fp16)[name = tensor("inputs_101_cast_fp16")]; + tensor var_3951 = const()[name = tensor("op_3951"), val = tensor(3)]; + tensor out_101_axes_0 = const()[name = tensor("out_101_axes_0"), val = tensor([1])]; + tensor var_3982_to_fp16 = const()[name = tensor("op_3982_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_101_cast_fp16 = layer_norm(axes = out_101_axes_0, epsilon = var_3982_to_fp16, x = inputs_101_cast_fp16)[name = tensor("out_101_cast_fp16")]; + tensor input_277_gamma_0_to_fp16 = const()[name = tensor("input_277_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235153664)))]; + tensor input_277_beta_0_to_fp16 = const()[name = tensor("input_277_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235155776)))]; + tensor input_277_epsilon_0_to_fp16 = const()[name = tensor("input_277_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_277_cast_fp16 = batch_norm(beta = input_277_beta_0_to_fp16, epsilon = input_277_epsilon_0_to_fp16, gamma = input_277_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_101_cast_fp16)[name = tensor("input_277_cast_fp16")]; + tensor var_4002_pad_type_0 = const()[name = tensor("op_4002_pad_type_0"), val = tensor("valid")]; + tensor var_4002_strides_0 = const()[name = tensor("op_4002_strides_0"), val = tensor([1, 1])]; + tensor var_4002_pad_0 = const()[name = tensor("op_4002_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4002_dilations_0 = const()[name = tensor("op_4002_dilations_0"), val = tensor([1, 1])]; + tensor var_4002_groups_0 = const()[name = tensor("op_4002_groups_0"), val = tensor(1)]; + tensor layers_10_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(235157888))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238303680))), name = tensor("layers_10_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_4002_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_4002_dilations_0, groups = var_4002_groups_0, pad = var_4002_pad_0, pad_type = var_4002_pad_type_0, strides = var_4002_strides_0, weight = layers_10_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_277_cast_fp16)[name = tensor("op_4002_cast_fp16")]; + tensor var_4008_pad_type_0 = const()[name = tensor("op_4008_pad_type_0"), val = tensor("valid")]; + tensor var_4008_strides_0 = const()[name = tensor("op_4008_strides_0"), val = tensor([1, 1])]; + tensor var_4008_pad_0 = const()[name = tensor("op_4008_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4008_dilations_0 = const()[name = tensor("op_4008_dilations_0"), val = tensor([1, 1])]; + tensor var_4008_groups_0 = const()[name = tensor("op_4008_groups_0"), val = tensor(1)]; + tensor layers_10_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238476416))), name = tensor("layers_10_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238303872))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_4008_cast_fp16 = conv(dilations = var_4008_dilations_0, groups = var_4008_groups_0, pad = var_4008_pad_0, pad_type = var_4008_pad_type_0, strides = var_4008_strides_0, weight = layers_10_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_277_cast_fp16)[name = tensor("op_4008_cast_fp16")]; + tensor input_279_cast_fp16 = add(x = var_4002_cast_fp16, y = var_4008_cast_fp16)[name = tensor("input_279_cast_fp16")]; + tensor input_281_cast_fp16 = silu(x = input_279_cast_fp16)[name = tensor("input_281_cast_fp16")]; + tensor var_4019_pad_type_0 = const()[name = tensor("op_4019_pad_type_0"), val = tensor("valid")]; + tensor var_4019_strides_0 = const()[name = tensor("op_4019_strides_0"), val = tensor([1, 1])]; + tensor var_4019_pad_0 = const()[name = tensor("op_4019_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4019_dilations_0 = const()[name = tensor("op_4019_dilations_0"), val = tensor([1, 1])]; + tensor var_4019_groups_0 = const()[name = tensor("op_4019_groups_0"), val = tensor(1)]; + tensor layers_10_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(239000768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242146560))), name = tensor("layers_10_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_4019_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4019_dilations_0, groups = var_4019_groups_0, pad = var_4019_pad_0, pad_type = var_4019_pad_type_0, strides = var_4019_strides_0, weight = layers_10_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_281_cast_fp16)[name = tensor("op_4019_cast_fp16")]; + tensor var_4025_pad_type_0 = const()[name = tensor("op_4025_pad_type_0"), val = tensor("valid")]; + tensor var_4025_strides_0 = const()[name = tensor("op_4025_strides_0"), val = tensor([1, 1])]; + tensor var_4025_pad_0 = const()[name = tensor("op_4025_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4025_dilations_0 = const()[name = tensor("op_4025_dilations_0"), val = tensor([1, 1])]; + tensor var_4025_groups_0 = const()[name = tensor("op_4025_groups_0"), val = tensor(1)]; + tensor layers_10_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242300032))), name = tensor("layers_10_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242146752))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_4025_cast_fp16 = conv(dilations = var_4025_dilations_0, groups = var_4025_groups_0, pad = var_4025_pad_0, pad_type = var_4025_pad_type_0, strides = var_4025_strides_0, weight = layers_10_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_281_cast_fp16)[name = tensor("op_4025_cast_fp16")]; + tensor x_63_cast_fp16 = add(x = var_4019_cast_fp16, y = var_4025_cast_fp16)[name = tensor("x_63_cast_fp16")]; + tensor var_4027_to_fp16 = const()[name = tensor("op_4027_to_fp16"), val = tensor(0x1p-1)]; + tensor var_4028_cast_fp16 = mul(x = x_63_cast_fp16, y = var_4027_to_fp16)[name = tensor("op_4028_cast_fp16")]; + tensor inputs_103_cast_fp16 = add(x = inputs_101_cast_fp16, y = var_4028_cast_fp16)[name = tensor("inputs_103_cast_fp16")]; + tensor out_103_axes_0 = const()[name = tensor("out_103_axes_0"), val = tensor([1])]; + tensor var_4038_to_fp16 = const()[name = tensor("op_4038_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_103_cast_fp16 = layer_norm(axes = out_103_axes_0, epsilon = var_4038_to_fp16, x = inputs_103_cast_fp16)[name = tensor("out_103_cast_fp16")]; + tensor obj_43_gamma_0_to_fp16 = const()[name = tensor("obj_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242824384)))]; + tensor obj_43_beta_0_to_fp16 = const()[name = tensor("obj_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242826496)))]; + tensor obj_43_epsilon_0_to_fp16 = const()[name = tensor("obj_43_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_43_cast_fp16 = batch_norm(beta = obj_43_beta_0_to_fp16, epsilon = obj_43_epsilon_0_to_fp16, gamma = obj_43_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_103_cast_fp16)[name = tensor("obj_43_cast_fp16")]; + tensor var_4063_pad_type_0 = const()[name = tensor("op_4063_pad_type_0"), val = tensor("valid")]; + tensor var_4063_strides_0 = const()[name = tensor("op_4063_strides_0"), val = tensor([1, 1])]; + tensor var_4063_pad_0 = const()[name = tensor("op_4063_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4063_dilations_0 = const()[name = tensor("op_4063_dilations_0"), val = tensor([1, 1])]; + tensor var_4063_groups_0 = const()[name = tensor("op_4063_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242828608))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(243615104))), name = tensor("layers_10_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4063_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4063_dilations_0, groups = var_4063_groups_0, pad = var_4063_pad_0, pad_type = var_4063_pad_type_0, strides = var_4063_strides_0, weight = layers_10_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_43_cast_fp16)[name = tensor("op_4063_cast_fp16")]; + tensor var_4069_pad_type_0 = const()[name = tensor("op_4069_pad_type_0"), val = tensor("valid")]; + tensor var_4069_strides_0 = const()[name = tensor("op_4069_strides_0"), val = tensor([1, 1])]; + tensor var_4069_pad_0 = const()[name = tensor("op_4069_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4069_dilations_0 = const()[name = tensor("op_4069_dilations_0"), val = tensor([1, 1])]; + tensor var_4069_groups_0 = const()[name = tensor("op_4069_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(243651328))), name = tensor("layers_10_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(243615296))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4069_cast_fp16 = conv(dilations = var_4069_dilations_0, groups = var_4069_groups_0, pad = var_4069_pad_0, pad_type = var_4069_pad_type_0, strides = var_4069_strides_0, weight = layers_10_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_43_cast_fp16)[name = tensor("op_4069_cast_fp16")]; + tensor query_41_cast_fp16 = add(x = var_4063_cast_fp16, y = var_4069_cast_fp16)[name = tensor("query_41_cast_fp16")]; + tensor var_4078_pad_type_0 = const()[name = tensor("op_4078_pad_type_0"), val = tensor("valid")]; + tensor var_4078_strides_0 = const()[name = tensor("op_4078_strides_0"), val = tensor([1, 1])]; + tensor var_4078_pad_0 = const()[name = tensor("op_4078_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4078_dilations_0 = const()[name = tensor("op_4078_dilations_0"), val = tensor([1, 1])]; + tensor var_4078_groups_0 = const()[name = tensor("op_4078_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(243782464))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(244568960))), name = tensor("layers_10_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4078_cast_fp16 = conv(dilations = var_4078_dilations_0, groups = var_4078_groups_0, pad = var_4078_pad_0, pad_type = var_4078_pad_type_0, strides = var_4078_strides_0, weight = layers_10_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_43_cast_fp16)[name = tensor("op_4078_cast_fp16")]; + tensor var_4084_pad_type_0 = const()[name = tensor("op_4084_pad_type_0"), val = tensor("valid")]; + tensor var_4084_strides_0 = const()[name = tensor("op_4084_strides_0"), val = tensor([1, 1])]; + tensor var_4084_pad_0 = const()[name = tensor("op_4084_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4084_dilations_0 = const()[name = tensor("op_4084_dilations_0"), val = tensor([1, 1])]; + tensor var_4084_groups_0 = const()[name = tensor("op_4084_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(244612096))), name = tensor("layers_10_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(244569152))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4084_cast_fp16 = conv(dilations = var_4084_dilations_0, groups = var_4084_groups_0, pad = var_4084_pad_0, pad_type = var_4084_pad_type_0, strides = var_4084_strides_0, weight = layers_10_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_43_cast_fp16)[name = tensor("op_4084_cast_fp16")]; + tensor key_21_cast_fp16 = add(x = var_4078_cast_fp16, y = var_4084_cast_fp16)[name = tensor("key_21_cast_fp16")]; + tensor var_4094_pad_type_0 = const()[name = tensor("op_4094_pad_type_0"), val = tensor("valid")]; + tensor var_4094_strides_0 = const()[name = tensor("op_4094_strides_0"), val = tensor([1, 1])]; + tensor var_4094_pad_0 = const()[name = tensor("op_4094_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4094_dilations_0 = const()[name = tensor("op_4094_dilations_0"), val = tensor([1, 1])]; + tensor var_4094_groups_0 = const()[name = tensor("op_4094_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(244743232))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(245529728))), name = tensor("layers_10_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4094_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4094_dilations_0, groups = var_4094_groups_0, pad = var_4094_pad_0, pad_type = var_4094_pad_type_0, strides = var_4094_strides_0, weight = layers_10_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_43_cast_fp16)[name = tensor("op_4094_cast_fp16")]; + tensor var_4100_pad_type_0 = const()[name = tensor("op_4100_pad_type_0"), val = tensor("valid")]; + tensor var_4100_strides_0 = const()[name = tensor("op_4100_strides_0"), val = tensor([1, 1])]; + tensor var_4100_pad_0 = const()[name = tensor("op_4100_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4100_dilations_0 = const()[name = tensor("op_4100_dilations_0"), val = tensor([1, 1])]; + tensor var_4100_groups_0 = const()[name = tensor("op_4100_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(245561088))), name = tensor("layers_10_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(245529920))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4100_cast_fp16 = conv(dilations = var_4100_dilations_0, groups = var_4100_groups_0, pad = var_4100_pad_0, pad_type = var_4100_pad_type_0, strides = var_4100_strides_0, weight = layers_10_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_43_cast_fp16)[name = tensor("op_4100_cast_fp16")]; + tensor value_21_cast_fp16 = add(x = var_4094_cast_fp16, y = var_4100_cast_fp16)[name = tensor("value_21_cast_fp16")]; + tensor var_4103_to_fp16 = const()[name = tensor("op_4103_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(245692224)))]; + tensor query_43_cast_fp16 = add(x = query_41_cast_fp16, y = var_4103_to_fp16)[name = tensor("query_43_cast_fp16")]; + tensor var_4106_to_fp16 = const()[name = tensor("op_4106_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(245694336)))]; + tensor q_with_bias_v_21_cast_fp16 = add(x = query_41_cast_fp16, y = var_4106_to_fp16)[name = tensor("q_with_bias_v_21_cast_fp16")]; + tensor var_4116_pad_type_0 = const()[name = tensor("op_4116_pad_type_0"), val = tensor("valid")]; + tensor var_4116_strides_0 = const()[name = tensor("op_4116_strides_0"), val = tensor([1, 1])]; + tensor var_4116_pad_0 = const()[name = tensor("op_4116_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4116_dilations_0 = const()[name = tensor("op_4116_dilations_0"), val = tensor([1, 1])]; + tensor var_4116_groups_0 = const()[name = tensor("op_4116_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(245696448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(246482944))), name = tensor("layers_10_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4116_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4116_dilations_0, groups = var_4116_groups_0, pad = var_4116_pad_0, pad_type = var_4116_pad_type_0, strides = var_4116_strides_0, weight = layers_10_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_4116_cast_fp16")]; + tensor var_4122_pad_type_0 = const()[name = tensor("op_4122_pad_type_0"), val = tensor("valid")]; + tensor var_4122_strides_0 = const()[name = tensor("op_4122_strides_0"), val = tensor([1, 1])]; + tensor var_4122_pad_0 = const()[name = tensor("op_4122_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4122_dilations_0 = const()[name = tensor("op_4122_dilations_0"), val = tensor([1, 1])]; + tensor var_4122_groups_0 = const()[name = tensor("op_4122_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(246539136))), name = tensor("layers_10_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(246483136))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4122_cast_fp16 = conv(dilations = var_4122_dilations_0, groups = var_4122_groups_0, pad = var_4122_pad_0, pad_type = var_4122_pad_type_0, strides = var_4122_strides_0, weight = layers_10_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_4122_cast_fp16")]; + tensor p_21_cast_fp16 = add(x = var_4116_cast_fp16, y = var_4122_cast_fp16)[name = tensor("p_21_cast_fp16")]; + tensor var_4126 = const()[name = tensor("op_4126"), val = tensor([1, 8, 128, 188])]; + tensor var_4127_cast_fp16 = reshape(shape = var_4126, x = q_with_bias_v_21_cast_fp16)[name = tensor("op_4127_cast_fp16")]; + tensor var_4128 = const()[name = tensor("op_4128"), val = tensor([1, 8, 128, -1])]; + tensor var_4129_cast_fp16 = reshape(shape = var_4128, x = p_21_cast_fp16)[name = tensor("op_4129_cast_fp16")]; + tensor matrix_bd_81_transpose_x_0 = const()[name = tensor("matrix_bd_81_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_81_transpose_y_0 = const()[name = tensor("matrix_bd_81_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_81_cast_fp16 = matmul(transpose_x = matrix_bd_81_transpose_x_0, transpose_y = matrix_bd_81_transpose_y_0, x = var_4127_cast_fp16, y = var_4129_cast_fp16)[name = tensor("matrix_bd_81_cast_fp16")]; + tensor matrix_bd_83_pad_0 = const()[name = tensor("matrix_bd_83_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_83_mode_0 = const()[name = tensor("matrix_bd_83_mode_0"), val = tensor("constant")]; + tensor const_120_to_fp16 = const()[name = tensor("const_120_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_83_cast_fp16 = pad(constant_val = const_120_to_fp16, mode = matrix_bd_83_mode_0, pad = matrix_bd_83_pad_0, x = matrix_bd_81_cast_fp16)[name = tensor("matrix_bd_83_cast_fp16")]; + tensor var_4138 = const()[name = tensor("op_4138"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_85_cast_fp16 = reshape(shape = var_4138, x = matrix_bd_83_cast_fp16)[name = tensor("matrix_bd_85_cast_fp16")]; + tensor var_4142_begin_0 = const()[name = tensor("op_4142_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_4142_end_0 = const()[name = tensor("op_4142_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_4142_end_mask_0 = const()[name = tensor("op_4142_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4142_cast_fp16 = slice_by_index(begin = var_4142_begin_0, end = var_4142_end_0, end_mask = var_4142_end_mask_0, x = matrix_bd_85_cast_fp16)[name = tensor("op_4142_cast_fp16")]; + tensor var_4143 = const()[name = tensor("op_4143"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_87_cast_fp16 = reshape(shape = var_4143, x = var_4142_cast_fp16)[name = tensor("matrix_bd_87_cast_fp16")]; + tensor var_4148_begin_0 = const()[name = tensor("op_4148_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4148_end_0 = const()[name = tensor("op_4148_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_4148_end_mask_0 = const()[name = tensor("op_4148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4148_cast_fp16 = slice_by_index(begin = var_4148_begin_0, end = var_4148_end_0, end_mask = var_4148_end_mask_0, x = matrix_bd_87_cast_fp16)[name = tensor("op_4148_cast_fp16")]; + tensor var_4149_to_fp16 = const()[name = tensor("op_4149_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_21_cast_fp16 = mul(x = var_4148_cast_fp16, y = var_4149_to_fp16)[name = tensor("qk_mask_21_cast_fp16")]; + tensor var_4153 = const()[name = tensor("op_4153"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_21_cast_fp16 = reshape(shape = var_4153, x = query_43_cast_fp16)[name = tensor("mh_q_21_cast_fp16")]; + tensor var_4155_to_fp16 = const()[name = tensor("op_4155_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_4156_cast_fp16 = mul(x = mh_q_21_cast_fp16, y = var_4155_to_fp16)[name = tensor("op_4156_cast_fp16")]; + tensor var_4159 = const()[name = tensor("op_4159"), val = tensor([1, 8, 128, 188])]; + tensor var_4160_cast_fp16 = reshape(shape = var_4159, x = key_21_cast_fp16)[name = tensor("op_4160_cast_fp16")]; + tensor mh_w_41_transpose_x_0 = const()[name = tensor("mh_w_41_transpose_x_0"), val = tensor(true)]; + tensor mh_w_41_transpose_y_0 = const()[name = tensor("mh_w_41_transpose_y_0"), val = tensor(false)]; + tensor mh_w_41_cast_fp16 = matmul(transpose_x = mh_w_41_transpose_x_0, transpose_y = mh_w_41_transpose_y_0, x = var_4156_cast_fp16, y = var_4160_cast_fp16)[name = tensor("mh_w_41_cast_fp16")]; + tensor mh_w_43_cast_fp16 = add(x = mh_w_41_cast_fp16, y = qk_mask_21_cast_fp16)[name = tensor("mh_w_43_cast_fp16")]; + tensor var_4164_cast_fp16 = softmax(axis = var_3951, x = mh_w_43_cast_fp16)[name = tensor("op_4164_cast_fp16")]; + tensor var_4165 = const()[name = tensor("op_4165"), val = tensor([1, 8, 128, 188])]; + tensor var_4166_cast_fp16 = reshape(shape = var_4165, x = value_21_cast_fp16)[name = tensor("op_4166_cast_fp16")]; + tensor attn_21_transpose_x_0 = const()[name = tensor("attn_21_transpose_x_0"), val = tensor(false)]; + tensor attn_21_transpose_y_0 = const()[name = tensor("attn_21_transpose_y_0"), val = tensor(true)]; + tensor attn_21_cast_fp16 = matmul(transpose_x = attn_21_transpose_x_0, transpose_y = attn_21_transpose_y_0, x = var_4166_cast_fp16, y = var_4164_cast_fp16)[name = tensor("attn_21_cast_fp16")]; + tensor var_4169 = const()[name = tensor("op_4169"), val = tensor([1, 1024, 1, 188])]; + tensor input_283_cast_fp16 = reshape(shape = var_4169, x = attn_21_cast_fp16)[name = tensor("input_283_cast_fp16")]; + tensor var_4179_pad_type_0 = const()[name = tensor("op_4179_pad_type_0"), val = tensor("valid")]; + tensor var_4179_strides_0 = const()[name = tensor("op_4179_strides_0"), val = tensor([1, 1])]; + tensor var_4179_pad_0 = const()[name = tensor("op_4179_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4179_dilations_0 = const()[name = tensor("op_4179_dilations_0"), val = tensor([1, 1])]; + tensor var_4179_groups_0 = const()[name = tensor("op_4179_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(246670272))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247456768))), name = tensor("layers_10_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4179_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4179_dilations_0, groups = var_4179_groups_0, pad = var_4179_pad_0, pad_type = var_4179_pad_type_0, strides = var_4179_strides_0, weight = layers_10_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_283_cast_fp16)[name = tensor("op_4179_cast_fp16")]; + tensor var_4185_pad_type_0 = const()[name = tensor("op_4185_pad_type_0"), val = tensor("valid")]; + tensor var_4185_strides_0 = const()[name = tensor("op_4185_strides_0"), val = tensor([1, 1])]; + tensor var_4185_pad_0 = const()[name = tensor("op_4185_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4185_dilations_0 = const()[name = tensor("op_4185_dilations_0"), val = tensor([1, 1])]; + tensor var_4185_groups_0 = const()[name = tensor("op_4185_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247489024))), name = tensor("layers_10_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247456960))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4185_cast_fp16 = conv(dilations = var_4185_dilations_0, groups = var_4185_groups_0, pad = var_4185_pad_0, pad_type = var_4185_pad_type_0, strides = var_4185_strides_0, weight = layers_10_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_283_cast_fp16)[name = tensor("op_4185_cast_fp16")]; + tensor obj_45_cast_fp16 = add(x = var_4179_cast_fp16, y = var_4185_cast_fp16)[name = tensor("obj_45_cast_fp16")]; + tensor inputs_105_cast_fp16 = add(x = inputs_103_cast_fp16, y = obj_45_cast_fp16)[name = tensor("inputs_105_cast_fp16")]; + tensor out_105_axes_0 = const()[name = tensor("out_105_axes_0"), val = tensor([1])]; + tensor var_4196_to_fp16 = const()[name = tensor("op_4196_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_105_cast_fp16 = layer_norm(axes = out_105_axes_0, epsilon = var_4196_to_fp16, x = inputs_105_cast_fp16)[name = tensor("out_105_cast_fp16")]; + tensor input_285_gamma_0_to_fp16 = const()[name = tensor("input_285_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247620160)))]; + tensor input_285_beta_0_to_fp16 = const()[name = tensor("input_285_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247622272)))]; + tensor input_285_epsilon_0_to_fp16 = const()[name = tensor("input_285_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_285_cast_fp16 = batch_norm(beta = input_285_beta_0_to_fp16, epsilon = input_285_epsilon_0_to_fp16, gamma = input_285_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_105_cast_fp16)[name = tensor("input_285_cast_fp16")]; + tensor var_4217_pad_type_0 = const()[name = tensor("op_4217_pad_type_0"), val = tensor("valid")]; + tensor var_4217_strides_0 = const()[name = tensor("op_4217_strides_0"), val = tensor([1, 1])]; + tensor var_4217_pad_0 = const()[name = tensor("op_4217_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4217_dilations_0 = const()[name = tensor("op_4217_dilations_0"), val = tensor([1, 1])]; + tensor var_4217_groups_0 = const()[name = tensor("op_4217_groups_0"), val = tensor(1)]; + tensor layers_10_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(247624384))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249197312))), name = tensor("layers_10_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_4217_cast_fp16 = conv(dilations = var_4217_dilations_0, groups = var_4217_groups_0, pad = var_4217_pad_0, pad_type = var_4217_pad_type_0, strides = var_4217_strides_0, weight = layers_10_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_285_cast_fp16)[name = tensor("op_4217_cast_fp16")]; + tensor var_4223_pad_type_0 = const()[name = tensor("op_4223_pad_type_0"), val = tensor("valid")]; + tensor var_4223_strides_0 = const()[name = tensor("op_4223_strides_0"), val = tensor([1, 1])]; + tensor var_4223_pad_0 = const()[name = tensor("op_4223_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4223_dilations_0 = const()[name = tensor("op_4223_dilations_0"), val = tensor([1, 1])]; + tensor var_4223_groups_0 = const()[name = tensor("op_4223_groups_0"), val = tensor(1)]; + tensor layers_10_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249257856))), name = tensor("layers_10_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249197504))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_4223_cast_fp16 = conv(dilations = var_4223_dilations_0, groups = var_4223_groups_0, pad = var_4223_pad_0, pad_type = var_4223_pad_type_0, strides = var_4223_strides_0, weight = layers_10_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_285_cast_fp16)[name = tensor("op_4223_cast_fp16")]; + tensor input_287_cast_fp16 = add(x = var_4217_cast_fp16, y = var_4223_cast_fp16)[name = tensor("input_287_cast_fp16")]; + tensor input_289_split_num_splits_0 = const()[name = tensor("input_289_split_num_splits_0"), val = tensor(2)]; + tensor input_289_split_axis_0 = const()[name = tensor("input_289_split_axis_0"), val = tensor(1)]; + tensor input_289_split_cast_fp16_0, tensor input_289_split_cast_fp16_1 = split(axis = input_289_split_axis_0, num_splits = input_289_split_num_splits_0, x = input_287_cast_fp16)[name = tensor("input_289_split_cast_fp16")]; + tensor input_289_split_1_sigmoid_cast_fp16 = sigmoid(x = input_289_split_cast_fp16_1)[name = tensor("input_289_split_1_sigmoid_cast_fp16")]; + tensor input_289_cast_fp16 = mul(x = input_289_split_cast_fp16_0, y = input_289_split_1_sigmoid_cast_fp16)[name = tensor("input_289_cast_fp16")]; + tensor input_291_pad_type_0 = const()[name = tensor("input_291_pad_type_0"), val = tensor("custom")]; + tensor input_291_pad_0 = const()[name = tensor("input_291_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_291_groups_0 = const()[name = tensor("input_291_groups_0"), val = tensor(1024)]; + tensor input_291_strides_0 = const()[name = tensor("input_291_strides_0"), val = tensor([1, 1])]; + tensor input_291_dilations_0 = const()[name = tensor("input_291_dilations_0"), val = tensor([1, 1])]; + tensor const_288_to_fp16 = const()[name = tensor("const_288_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249520064)))]; + tensor const_289_to_fp16 = const()[name = tensor("const_289_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249538560)))]; + tensor input_293_cast_fp16 = conv(bias = const_289_to_fp16, dilations = input_291_dilations_0, groups = input_291_groups_0, pad = input_291_pad_0, pad_type = input_291_pad_type_0, strides = input_291_strides_0, weight = const_288_to_fp16, x = input_289_cast_fp16)[name = tensor("input_293_cast_fp16")]; + tensor input_295_cast_fp16 = silu(x = input_293_cast_fp16)[name = tensor("input_295_cast_fp16")]; + tensor var_4245_pad_type_0 = const()[name = tensor("op_4245_pad_type_0"), val = tensor("valid")]; + tensor var_4245_strides_0 = const()[name = tensor("op_4245_strides_0"), val = tensor([1, 1])]; + tensor var_4245_pad_0 = const()[name = tensor("op_4245_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4245_dilations_0 = const()[name = tensor("op_4245_dilations_0"), val = tensor([1, 1])]; + tensor var_4245_groups_0 = const()[name = tensor("op_4245_groups_0"), val = tensor(1)]; + tensor layers_10_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(249540672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250327168))), name = tensor("layers_10_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4245_cast_fp16 = conv(dilations = var_4245_dilations_0, groups = var_4245_groups_0, pad = var_4245_pad_0, pad_type = var_4245_pad_type_0, strides = var_4245_strides_0, weight = layers_10_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_295_cast_fp16)[name = tensor("op_4245_cast_fp16")]; + tensor var_4251_pad_type_0 = const()[name = tensor("op_4251_pad_type_0"), val = tensor("valid")]; + tensor var_4251_strides_0 = const()[name = tensor("op_4251_strides_0"), val = tensor([1, 1])]; + tensor var_4251_pad_0 = const()[name = tensor("op_4251_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4251_dilations_0 = const()[name = tensor("op_4251_dilations_0"), val = tensor([1, 1])]; + tensor var_4251_groups_0 = const()[name = tensor("op_4251_groups_0"), val = tensor(1)]; + tensor layers_10_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250358720))), name = tensor("layers_10_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250327360))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4251_cast_fp16 = conv(dilations = var_4251_dilations_0, groups = var_4251_groups_0, pad = var_4251_pad_0, pad_type = var_4251_pad_type_0, strides = var_4251_strides_0, weight = layers_10_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_295_cast_fp16)[name = tensor("op_4251_cast_fp16")]; + tensor x_65_cast_fp16 = add(x = var_4245_cast_fp16, y = var_4251_cast_fp16)[name = tensor("x_65_cast_fp16")]; + tensor inputs_107_cast_fp16 = add(x = inputs_105_cast_fp16, y = x_65_cast_fp16)[name = tensor("inputs_107_cast_fp16")]; + tensor out_107_axes_0 = const()[name = tensor("out_107_axes_0"), val = tensor([1])]; + tensor var_4262_to_fp16 = const()[name = tensor("op_4262_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_107_cast_fp16 = layer_norm(axes = out_107_axes_0, epsilon = var_4262_to_fp16, x = inputs_107_cast_fp16)[name = tensor("out_107_cast_fp16")]; + tensor input_297_gamma_0_to_fp16 = const()[name = tensor("input_297_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250489856)))]; + tensor input_297_beta_0_to_fp16 = const()[name = tensor("input_297_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250491968)))]; + tensor input_297_epsilon_0_to_fp16 = const()[name = tensor("input_297_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_297_cast_fp16 = batch_norm(beta = input_297_beta_0_to_fp16, epsilon = input_297_epsilon_0_to_fp16, gamma = input_297_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_107_cast_fp16)[name = tensor("input_297_cast_fp16")]; + tensor var_4282_pad_type_0 = const()[name = tensor("op_4282_pad_type_0"), val = tensor("valid")]; + tensor var_4282_strides_0 = const()[name = tensor("op_4282_strides_0"), val = tensor([1, 1])]; + tensor var_4282_pad_0 = const()[name = tensor("op_4282_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4282_dilations_0 = const()[name = tensor("op_4282_dilations_0"), val = tensor([1, 1])]; + tensor var_4282_groups_0 = const()[name = tensor("op_4282_groups_0"), val = tensor(1)]; + tensor layers_10_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250494080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253639872))), name = tensor("layers_10_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_4282_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_4282_dilations_0, groups = var_4282_groups_0, pad = var_4282_pad_0, pad_type = var_4282_pad_type_0, strides = var_4282_strides_0, weight = layers_10_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_297_cast_fp16)[name = tensor("op_4282_cast_fp16")]; + tensor var_4288_pad_type_0 = const()[name = tensor("op_4288_pad_type_0"), val = tensor("valid")]; + tensor var_4288_strides_0 = const()[name = tensor("op_4288_strides_0"), val = tensor([1, 1])]; + tensor var_4288_pad_0 = const()[name = tensor("op_4288_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4288_dilations_0 = const()[name = tensor("op_4288_dilations_0"), val = tensor([1, 1])]; + tensor var_4288_groups_0 = const()[name = tensor("op_4288_groups_0"), val = tensor(1)]; + tensor layers_10_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253796736))), name = tensor("layers_10_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253640064))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_4288_cast_fp16 = conv(dilations = var_4288_dilations_0, groups = var_4288_groups_0, pad = var_4288_pad_0, pad_type = var_4288_pad_type_0, strides = var_4288_strides_0, weight = layers_10_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_297_cast_fp16)[name = tensor("op_4288_cast_fp16")]; + tensor input_299_cast_fp16 = add(x = var_4282_cast_fp16, y = var_4288_cast_fp16)[name = tensor("input_299_cast_fp16")]; + tensor input_301_cast_fp16 = silu(x = input_299_cast_fp16)[name = tensor("input_301_cast_fp16")]; + tensor var_4299_pad_type_0 = const()[name = tensor("op_4299_pad_type_0"), val = tensor("valid")]; + tensor var_4299_strides_0 = const()[name = tensor("op_4299_strides_0"), val = tensor([1, 1])]; + tensor var_4299_pad_0 = const()[name = tensor("op_4299_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4299_dilations_0 = const()[name = tensor("op_4299_dilations_0"), val = tensor([1, 1])]; + tensor var_4299_groups_0 = const()[name = tensor("op_4299_groups_0"), val = tensor(1)]; + tensor layers_10_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(254321088))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(257466880))), name = tensor("layers_10_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_4299_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4299_dilations_0, groups = var_4299_groups_0, pad = var_4299_pad_0, pad_type = var_4299_pad_type_0, strides = var_4299_strides_0, weight = layers_10_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_301_cast_fp16)[name = tensor("op_4299_cast_fp16")]; + tensor var_4305_pad_type_0 = const()[name = tensor("op_4305_pad_type_0"), val = tensor("valid")]; + tensor var_4305_strides_0 = const()[name = tensor("op_4305_strides_0"), val = tensor([1, 1])]; + tensor var_4305_pad_0 = const()[name = tensor("op_4305_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4305_dilations_0 = const()[name = tensor("op_4305_dilations_0"), val = tensor([1, 1])]; + tensor var_4305_groups_0 = const()[name = tensor("op_4305_groups_0"), val = tensor(1)]; + tensor layers_10_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(257617024))), name = tensor("layers_10_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(257467072))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_4305_cast_fp16 = conv(dilations = var_4305_dilations_0, groups = var_4305_groups_0, pad = var_4305_pad_0, pad_type = var_4305_pad_type_0, strides = var_4305_strides_0, weight = layers_10_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_301_cast_fp16)[name = tensor("op_4305_cast_fp16")]; + tensor x_67_cast_fp16 = add(x = var_4299_cast_fp16, y = var_4305_cast_fp16)[name = tensor("x_67_cast_fp16")]; + tensor var_4307_to_fp16 = const()[name = tensor("op_4307_to_fp16"), val = tensor(0x1p-1)]; + tensor var_4308_cast_fp16 = mul(x = x_67_cast_fp16, y = var_4307_to_fp16)[name = tensor("op_4308_cast_fp16")]; + tensor inputs_109_cast_fp16 = add(x = inputs_107_cast_fp16, y = var_4308_cast_fp16)[name = tensor("inputs_109_cast_fp16")]; + tensor out_109_axes_0 = const()[name = tensor("out_109_axes_0"), val = tensor([1])]; + tensor var_4318_to_fp16 = const()[name = tensor("op_4318_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_109_cast_fp16 = layer_norm(axes = out_109_axes_0, epsilon = var_4318_to_fp16, x = inputs_109_cast_fp16)[name = tensor("out_109_cast_fp16")]; + tensor inputs_111_gamma_0_to_fp16 = const()[name = tensor("inputs_111_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(258141376)))]; + tensor inputs_111_beta_0_to_fp16 = const()[name = tensor("inputs_111_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(258143488)))]; + tensor inputs_111_epsilon_0_to_fp16 = const()[name = tensor("inputs_111_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_111_cast_fp16 = batch_norm(beta = inputs_111_beta_0_to_fp16, epsilon = inputs_111_epsilon_0_to_fp16, gamma = inputs_111_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_109_cast_fp16)[name = tensor("inputs_111_cast_fp16")]; + tensor var_4332 = const()[name = tensor("op_4332"), val = tensor(3)]; + tensor out_111_axes_0 = const()[name = tensor("out_111_axes_0"), val = tensor([1])]; + tensor var_4363_to_fp16 = const()[name = tensor("op_4363_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_111_cast_fp16 = layer_norm(axes = out_111_axes_0, epsilon = var_4363_to_fp16, x = inputs_111_cast_fp16)[name = tensor("out_111_cast_fp16")]; + tensor input_303_gamma_0_to_fp16 = const()[name = tensor("input_303_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(258145600)))]; + tensor input_303_beta_0_to_fp16 = const()[name = tensor("input_303_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(258147712)))]; + tensor input_303_epsilon_0_to_fp16 = const()[name = tensor("input_303_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_303_cast_fp16 = batch_norm(beta = input_303_beta_0_to_fp16, epsilon = input_303_epsilon_0_to_fp16, gamma = input_303_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_111_cast_fp16)[name = tensor("input_303_cast_fp16")]; + tensor var_4383_pad_type_0 = const()[name = tensor("op_4383_pad_type_0"), val = tensor("valid")]; + tensor var_4383_strides_0 = const()[name = tensor("op_4383_strides_0"), val = tensor([1, 1])]; + tensor var_4383_pad_0 = const()[name = tensor("op_4383_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4383_dilations_0 = const()[name = tensor("op_4383_dilations_0"), val = tensor([1, 1])]; + tensor var_4383_groups_0 = const()[name = tensor("op_4383_groups_0"), val = tensor(1)]; + tensor layers_11_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(258149824))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261295616))), name = tensor("layers_11_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_4383_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_4383_dilations_0, groups = var_4383_groups_0, pad = var_4383_pad_0, pad_type = var_4383_pad_type_0, strides = var_4383_strides_0, weight = layers_11_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_303_cast_fp16)[name = tensor("op_4383_cast_fp16")]; + tensor var_4389_pad_type_0 = const()[name = tensor("op_4389_pad_type_0"), val = tensor("valid")]; + tensor var_4389_strides_0 = const()[name = tensor("op_4389_strides_0"), val = tensor([1, 1])]; + tensor var_4389_pad_0 = const()[name = tensor("op_4389_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4389_dilations_0 = const()[name = tensor("op_4389_dilations_0"), val = tensor([1, 1])]; + tensor var_4389_groups_0 = const()[name = tensor("op_4389_groups_0"), val = tensor(1)]; + tensor layers_11_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261464640))), name = tensor("layers_11_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261295808))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_4389_cast_fp16 = conv(dilations = var_4389_dilations_0, groups = var_4389_groups_0, pad = var_4389_pad_0, pad_type = var_4389_pad_type_0, strides = var_4389_strides_0, weight = layers_11_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_303_cast_fp16)[name = tensor("op_4389_cast_fp16")]; + tensor input_305_cast_fp16 = add(x = var_4383_cast_fp16, y = var_4389_cast_fp16)[name = tensor("input_305_cast_fp16")]; + tensor input_307_cast_fp16 = silu(x = input_305_cast_fp16)[name = tensor("input_307_cast_fp16")]; + tensor var_4400_pad_type_0 = const()[name = tensor("op_4400_pad_type_0"), val = tensor("valid")]; + tensor var_4400_strides_0 = const()[name = tensor("op_4400_strides_0"), val = tensor([1, 1])]; + tensor var_4400_pad_0 = const()[name = tensor("op_4400_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4400_dilations_0 = const()[name = tensor("op_4400_dilations_0"), val = tensor([1, 1])]; + tensor var_4400_groups_0 = const()[name = tensor("op_4400_groups_0"), val = tensor(1)]; + tensor layers_11_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261988992))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265134784))), name = tensor("layers_11_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_4400_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4400_dilations_0, groups = var_4400_groups_0, pad = var_4400_pad_0, pad_type = var_4400_pad_type_0, strides = var_4400_strides_0, weight = layers_11_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_307_cast_fp16)[name = tensor("op_4400_cast_fp16")]; + tensor var_4406_pad_type_0 = const()[name = tensor("op_4406_pad_type_0"), val = tensor("valid")]; + tensor var_4406_strides_0 = const()[name = tensor("op_4406_strides_0"), val = tensor([1, 1])]; + tensor var_4406_pad_0 = const()[name = tensor("op_4406_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4406_dilations_0 = const()[name = tensor("op_4406_dilations_0"), val = tensor([1, 1])]; + tensor var_4406_groups_0 = const()[name = tensor("op_4406_groups_0"), val = tensor(1)]; + tensor layers_11_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265286336))), name = tensor("layers_11_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265134976))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_4406_cast_fp16 = conv(dilations = var_4406_dilations_0, groups = var_4406_groups_0, pad = var_4406_pad_0, pad_type = var_4406_pad_type_0, strides = var_4406_strides_0, weight = layers_11_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_307_cast_fp16)[name = tensor("op_4406_cast_fp16")]; + tensor x_69_cast_fp16 = add(x = var_4400_cast_fp16, y = var_4406_cast_fp16)[name = tensor("x_69_cast_fp16")]; + tensor var_4408_to_fp16 = const()[name = tensor("op_4408_to_fp16"), val = tensor(0x1p-1)]; + tensor var_4409_cast_fp16 = mul(x = x_69_cast_fp16, y = var_4408_to_fp16)[name = tensor("op_4409_cast_fp16")]; + tensor inputs_113_cast_fp16 = add(x = inputs_111_cast_fp16, y = var_4409_cast_fp16)[name = tensor("inputs_113_cast_fp16")]; + tensor out_113_axes_0 = const()[name = tensor("out_113_axes_0"), val = tensor([1])]; + tensor var_4419_to_fp16 = const()[name = tensor("op_4419_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_113_cast_fp16 = layer_norm(axes = out_113_axes_0, epsilon = var_4419_to_fp16, x = inputs_113_cast_fp16)[name = tensor("out_113_cast_fp16")]; + tensor obj_47_gamma_0_to_fp16 = const()[name = tensor("obj_47_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265810688)))]; + tensor obj_47_beta_0_to_fp16 = const()[name = tensor("obj_47_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265812800)))]; + tensor obj_47_epsilon_0_to_fp16 = const()[name = tensor("obj_47_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_47_cast_fp16 = batch_norm(beta = obj_47_beta_0_to_fp16, epsilon = obj_47_epsilon_0_to_fp16, gamma = obj_47_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_113_cast_fp16)[name = tensor("obj_47_cast_fp16")]; + tensor var_4444_pad_type_0 = const()[name = tensor("op_4444_pad_type_0"), val = tensor("valid")]; + tensor var_4444_strides_0 = const()[name = tensor("op_4444_strides_0"), val = tensor([1, 1])]; + tensor var_4444_pad_0 = const()[name = tensor("op_4444_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4444_dilations_0 = const()[name = tensor("op_4444_dilations_0"), val = tensor([1, 1])]; + tensor var_4444_groups_0 = const()[name = tensor("op_4444_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265814912))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(266601408))), name = tensor("layers_11_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4444_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4444_dilations_0, groups = var_4444_groups_0, pad = var_4444_pad_0, pad_type = var_4444_pad_type_0, strides = var_4444_strides_0, weight = layers_11_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_47_cast_fp16)[name = tensor("op_4444_cast_fp16")]; + tensor var_4450_pad_type_0 = const()[name = tensor("op_4450_pad_type_0"), val = tensor("valid")]; + tensor var_4450_strides_0 = const()[name = tensor("op_4450_strides_0"), val = tensor([1, 1])]; + tensor var_4450_pad_0 = const()[name = tensor("op_4450_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4450_dilations_0 = const()[name = tensor("op_4450_dilations_0"), val = tensor([1, 1])]; + tensor var_4450_groups_0 = const()[name = tensor("op_4450_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(266632640))), name = tensor("layers_11_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(266601600))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4450_cast_fp16 = conv(dilations = var_4450_dilations_0, groups = var_4450_groups_0, pad = var_4450_pad_0, pad_type = var_4450_pad_type_0, strides = var_4450_strides_0, weight = layers_11_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_47_cast_fp16)[name = tensor("op_4450_cast_fp16")]; + tensor query_45_cast_fp16 = add(x = var_4444_cast_fp16, y = var_4450_cast_fp16)[name = tensor("query_45_cast_fp16")]; + tensor var_4459_pad_type_0 = const()[name = tensor("op_4459_pad_type_0"), val = tensor("valid")]; + tensor var_4459_strides_0 = const()[name = tensor("op_4459_strides_0"), val = tensor([1, 1])]; + tensor var_4459_pad_0 = const()[name = tensor("op_4459_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4459_dilations_0 = const()[name = tensor("op_4459_dilations_0"), val = tensor([1, 1])]; + tensor var_4459_groups_0 = const()[name = tensor("op_4459_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(266763776))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(267550272))), name = tensor("layers_11_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4459_cast_fp16 = conv(dilations = var_4459_dilations_0, groups = var_4459_groups_0, pad = var_4459_pad_0, pad_type = var_4459_pad_type_0, strides = var_4459_strides_0, weight = layers_11_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_47_cast_fp16)[name = tensor("op_4459_cast_fp16")]; + tensor var_4465_pad_type_0 = const()[name = tensor("op_4465_pad_type_0"), val = tensor("valid")]; + tensor var_4465_strides_0 = const()[name = tensor("op_4465_strides_0"), val = tensor([1, 1])]; + tensor var_4465_pad_0 = const()[name = tensor("op_4465_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4465_dilations_0 = const()[name = tensor("op_4465_dilations_0"), val = tensor([1, 1])]; + tensor var_4465_groups_0 = const()[name = tensor("op_4465_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(267584768))), name = tensor("layers_11_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(267550464))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4465_cast_fp16 = conv(dilations = var_4465_dilations_0, groups = var_4465_groups_0, pad = var_4465_pad_0, pad_type = var_4465_pad_type_0, strides = var_4465_strides_0, weight = layers_11_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_47_cast_fp16)[name = tensor("op_4465_cast_fp16")]; + tensor key_23_cast_fp16 = add(x = var_4459_cast_fp16, y = var_4465_cast_fp16)[name = tensor("key_23_cast_fp16")]; + tensor var_4475_pad_type_0 = const()[name = tensor("op_4475_pad_type_0"), val = tensor("valid")]; + tensor var_4475_strides_0 = const()[name = tensor("op_4475_strides_0"), val = tensor([1, 1])]; + tensor var_4475_pad_0 = const()[name = tensor("op_4475_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4475_dilations_0 = const()[name = tensor("op_4475_dilations_0"), val = tensor([1, 1])]; + tensor var_4475_groups_0 = const()[name = tensor("op_4475_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(267715904))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268502400))), name = tensor("layers_11_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4475_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4475_dilations_0, groups = var_4475_groups_0, pad = var_4475_pad_0, pad_type = var_4475_pad_type_0, strides = var_4475_strides_0, weight = layers_11_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_47_cast_fp16)[name = tensor("op_4475_cast_fp16")]; + tensor var_4481_pad_type_0 = const()[name = tensor("op_4481_pad_type_0"), val = tensor("valid")]; + tensor var_4481_strides_0 = const()[name = tensor("op_4481_strides_0"), val = tensor([1, 1])]; + tensor var_4481_pad_0 = const()[name = tensor("op_4481_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4481_dilations_0 = const()[name = tensor("op_4481_dilations_0"), val = tensor([1, 1])]; + tensor var_4481_groups_0 = const()[name = tensor("op_4481_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268534016))), name = tensor("layers_11_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268502592))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4481_cast_fp16 = conv(dilations = var_4481_dilations_0, groups = var_4481_groups_0, pad = var_4481_pad_0, pad_type = var_4481_pad_type_0, strides = var_4481_strides_0, weight = layers_11_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_47_cast_fp16)[name = tensor("op_4481_cast_fp16")]; + tensor value_23_cast_fp16 = add(x = var_4475_cast_fp16, y = var_4481_cast_fp16)[name = tensor("value_23_cast_fp16")]; + tensor var_4484_to_fp16 = const()[name = tensor("op_4484_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268665152)))]; + tensor query_47_cast_fp16 = add(x = query_45_cast_fp16, y = var_4484_to_fp16)[name = tensor("query_47_cast_fp16")]; + tensor var_4487_to_fp16 = const()[name = tensor("op_4487_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268667264)))]; + tensor q_with_bias_v_23_cast_fp16 = add(x = query_45_cast_fp16, y = var_4487_to_fp16)[name = tensor("q_with_bias_v_23_cast_fp16")]; + tensor var_4497_pad_type_0 = const()[name = tensor("op_4497_pad_type_0"), val = tensor("valid")]; + tensor var_4497_strides_0 = const()[name = tensor("op_4497_strides_0"), val = tensor([1, 1])]; + tensor var_4497_pad_0 = const()[name = tensor("op_4497_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4497_dilations_0 = const()[name = tensor("op_4497_dilations_0"), val = tensor([1, 1])]; + tensor var_4497_groups_0 = const()[name = tensor("op_4497_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268669376))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(269455872))), name = tensor("layers_11_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4497_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4497_dilations_0, groups = var_4497_groups_0, pad = var_4497_pad_0, pad_type = var_4497_pad_type_0, strides = var_4497_strides_0, weight = layers_11_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_4497_cast_fp16")]; + tensor var_4503_pad_type_0 = const()[name = tensor("op_4503_pad_type_0"), val = tensor("valid")]; + tensor var_4503_strides_0 = const()[name = tensor("op_4503_strides_0"), val = tensor([1, 1])]; + tensor var_4503_pad_0 = const()[name = tensor("op_4503_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4503_dilations_0 = const()[name = tensor("op_4503_dilations_0"), val = tensor([1, 1])]; + tensor var_4503_groups_0 = const()[name = tensor("op_4503_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(269514048))), name = tensor("layers_11_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(269456064))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4503_cast_fp16 = conv(dilations = var_4503_dilations_0, groups = var_4503_groups_0, pad = var_4503_pad_0, pad_type = var_4503_pad_type_0, strides = var_4503_strides_0, weight = layers_11_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_4503_cast_fp16")]; + tensor p_23_cast_fp16 = add(x = var_4497_cast_fp16, y = var_4503_cast_fp16)[name = tensor("p_23_cast_fp16")]; + tensor var_4507 = const()[name = tensor("op_4507"), val = tensor([1, 8, 128, 188])]; + tensor var_4508_cast_fp16 = reshape(shape = var_4507, x = q_with_bias_v_23_cast_fp16)[name = tensor("op_4508_cast_fp16")]; + tensor var_4509 = const()[name = tensor("op_4509"), val = tensor([1, 8, 128, -1])]; + tensor var_4510_cast_fp16 = reshape(shape = var_4509, x = p_23_cast_fp16)[name = tensor("op_4510_cast_fp16")]; + tensor matrix_bd_89_transpose_x_0 = const()[name = tensor("matrix_bd_89_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_89_transpose_y_0 = const()[name = tensor("matrix_bd_89_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_89_cast_fp16 = matmul(transpose_x = matrix_bd_89_transpose_x_0, transpose_y = matrix_bd_89_transpose_y_0, x = var_4508_cast_fp16, y = var_4510_cast_fp16)[name = tensor("matrix_bd_89_cast_fp16")]; + tensor matrix_bd_91_pad_0 = const()[name = tensor("matrix_bd_91_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_91_mode_0 = const()[name = tensor("matrix_bd_91_mode_0"), val = tensor("constant")]; + tensor const_131_to_fp16 = const()[name = tensor("const_131_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_91_cast_fp16 = pad(constant_val = const_131_to_fp16, mode = matrix_bd_91_mode_0, pad = matrix_bd_91_pad_0, x = matrix_bd_89_cast_fp16)[name = tensor("matrix_bd_91_cast_fp16")]; + tensor var_4519 = const()[name = tensor("op_4519"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_93_cast_fp16 = reshape(shape = var_4519, x = matrix_bd_91_cast_fp16)[name = tensor("matrix_bd_93_cast_fp16")]; + tensor var_4523_begin_0 = const()[name = tensor("op_4523_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_4523_end_0 = const()[name = tensor("op_4523_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_4523_end_mask_0 = const()[name = tensor("op_4523_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4523_cast_fp16 = slice_by_index(begin = var_4523_begin_0, end = var_4523_end_0, end_mask = var_4523_end_mask_0, x = matrix_bd_93_cast_fp16)[name = tensor("op_4523_cast_fp16")]; + tensor var_4524 = const()[name = tensor("op_4524"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_95_cast_fp16 = reshape(shape = var_4524, x = var_4523_cast_fp16)[name = tensor("matrix_bd_95_cast_fp16")]; + tensor var_4529_begin_0 = const()[name = tensor("op_4529_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4529_end_0 = const()[name = tensor("op_4529_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_4529_end_mask_0 = const()[name = tensor("op_4529_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4529_cast_fp16 = slice_by_index(begin = var_4529_begin_0, end = var_4529_end_0, end_mask = var_4529_end_mask_0, x = matrix_bd_95_cast_fp16)[name = tensor("op_4529_cast_fp16")]; + tensor var_4530_to_fp16 = const()[name = tensor("op_4530_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_23_cast_fp16 = mul(x = var_4529_cast_fp16, y = var_4530_to_fp16)[name = tensor("qk_mask_23_cast_fp16")]; + tensor var_4534 = const()[name = tensor("op_4534"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_23_cast_fp16 = reshape(shape = var_4534, x = query_47_cast_fp16)[name = tensor("mh_q_23_cast_fp16")]; + tensor var_4536_to_fp16 = const()[name = tensor("op_4536_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_4537_cast_fp16 = mul(x = mh_q_23_cast_fp16, y = var_4536_to_fp16)[name = tensor("op_4537_cast_fp16")]; + tensor var_4540 = const()[name = tensor("op_4540"), val = tensor([1, 8, 128, 188])]; + tensor var_4541_cast_fp16 = reshape(shape = var_4540, x = key_23_cast_fp16)[name = tensor("op_4541_cast_fp16")]; + tensor mh_w_45_transpose_x_0 = const()[name = tensor("mh_w_45_transpose_x_0"), val = tensor(true)]; + tensor mh_w_45_transpose_y_0 = const()[name = tensor("mh_w_45_transpose_y_0"), val = tensor(false)]; + tensor mh_w_45_cast_fp16 = matmul(transpose_x = mh_w_45_transpose_x_0, transpose_y = mh_w_45_transpose_y_0, x = var_4537_cast_fp16, y = var_4541_cast_fp16)[name = tensor("mh_w_45_cast_fp16")]; + tensor mh_w_47_cast_fp16 = add(x = mh_w_45_cast_fp16, y = qk_mask_23_cast_fp16)[name = tensor("mh_w_47_cast_fp16")]; + tensor var_4545_cast_fp16 = softmax(axis = var_4332, x = mh_w_47_cast_fp16)[name = tensor("op_4545_cast_fp16")]; + tensor var_4546 = const()[name = tensor("op_4546"), val = tensor([1, 8, 128, 188])]; + tensor var_4547_cast_fp16 = reshape(shape = var_4546, x = value_23_cast_fp16)[name = tensor("op_4547_cast_fp16")]; + tensor attn_23_transpose_x_0 = const()[name = tensor("attn_23_transpose_x_0"), val = tensor(false)]; + tensor attn_23_transpose_y_0 = const()[name = tensor("attn_23_transpose_y_0"), val = tensor(true)]; + tensor attn_23_cast_fp16 = matmul(transpose_x = attn_23_transpose_x_0, transpose_y = attn_23_transpose_y_0, x = var_4547_cast_fp16, y = var_4545_cast_fp16)[name = tensor("attn_23_cast_fp16")]; + tensor var_4550 = const()[name = tensor("op_4550"), val = tensor([1, 1024, 1, 188])]; + tensor input_309_cast_fp16 = reshape(shape = var_4550, x = attn_23_cast_fp16)[name = tensor("input_309_cast_fp16")]; + tensor var_4560_pad_type_0 = const()[name = tensor("op_4560_pad_type_0"), val = tensor("valid")]; + tensor var_4560_strides_0 = const()[name = tensor("op_4560_strides_0"), val = tensor([1, 1])]; + tensor var_4560_pad_0 = const()[name = tensor("op_4560_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4560_dilations_0 = const()[name = tensor("op_4560_dilations_0"), val = tensor([1, 1])]; + tensor var_4560_groups_0 = const()[name = tensor("op_4560_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(269645184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270431680))), name = tensor("layers_11_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4560_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4560_dilations_0, groups = var_4560_groups_0, pad = var_4560_pad_0, pad_type = var_4560_pad_type_0, strides = var_4560_strides_0, weight = layers_11_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_309_cast_fp16)[name = tensor("op_4560_cast_fp16")]; + tensor var_4566_pad_type_0 = const()[name = tensor("op_4566_pad_type_0"), val = tensor("valid")]; + tensor var_4566_strides_0 = const()[name = tensor("op_4566_strides_0"), val = tensor([1, 1])]; + tensor var_4566_pad_0 = const()[name = tensor("op_4566_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4566_dilations_0 = const()[name = tensor("op_4566_dilations_0"), val = tensor([1, 1])]; + tensor var_4566_groups_0 = const()[name = tensor("op_4566_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270464384))), name = tensor("layers_11_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270431872))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4566_cast_fp16 = conv(dilations = var_4566_dilations_0, groups = var_4566_groups_0, pad = var_4566_pad_0, pad_type = var_4566_pad_type_0, strides = var_4566_strides_0, weight = layers_11_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_309_cast_fp16)[name = tensor("op_4566_cast_fp16")]; + tensor obj_49_cast_fp16 = add(x = var_4560_cast_fp16, y = var_4566_cast_fp16)[name = tensor("obj_49_cast_fp16")]; + tensor inputs_115_cast_fp16 = add(x = inputs_113_cast_fp16, y = obj_49_cast_fp16)[name = tensor("inputs_115_cast_fp16")]; + tensor out_115_axes_0 = const()[name = tensor("out_115_axes_0"), val = tensor([1])]; + tensor var_4577_to_fp16 = const()[name = tensor("op_4577_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_115_cast_fp16 = layer_norm(axes = out_115_axes_0, epsilon = var_4577_to_fp16, x = inputs_115_cast_fp16)[name = tensor("out_115_cast_fp16")]; + tensor input_311_gamma_0_to_fp16 = const()[name = tensor("input_311_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270595520)))]; + tensor input_311_beta_0_to_fp16 = const()[name = tensor("input_311_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270597632)))]; + tensor input_311_epsilon_0_to_fp16 = const()[name = tensor("input_311_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_311_cast_fp16 = batch_norm(beta = input_311_beta_0_to_fp16, epsilon = input_311_epsilon_0_to_fp16, gamma = input_311_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_115_cast_fp16)[name = tensor("input_311_cast_fp16")]; + tensor var_4598_pad_type_0 = const()[name = tensor("op_4598_pad_type_0"), val = tensor("valid")]; + tensor var_4598_strides_0 = const()[name = tensor("op_4598_strides_0"), val = tensor([1, 1])]; + tensor var_4598_pad_0 = const()[name = tensor("op_4598_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4598_dilations_0 = const()[name = tensor("op_4598_dilations_0"), val = tensor([1, 1])]; + tensor var_4598_groups_0 = const()[name = tensor("op_4598_groups_0"), val = tensor(1)]; + tensor layers_11_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270599744))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272172672))), name = tensor("layers_11_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_4598_cast_fp16 = conv(dilations = var_4598_dilations_0, groups = var_4598_groups_0, pad = var_4598_pad_0, pad_type = var_4598_pad_type_0, strides = var_4598_strides_0, weight = layers_11_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_311_cast_fp16)[name = tensor("op_4598_cast_fp16")]; + tensor var_4604_pad_type_0 = const()[name = tensor("op_4604_pad_type_0"), val = tensor("valid")]; + tensor var_4604_strides_0 = const()[name = tensor("op_4604_strides_0"), val = tensor([1, 1])]; + tensor var_4604_pad_0 = const()[name = tensor("op_4604_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4604_dilations_0 = const()[name = tensor("op_4604_dilations_0"), val = tensor([1, 1])]; + tensor var_4604_groups_0 = const()[name = tensor("op_4604_groups_0"), val = tensor(1)]; + tensor layers_11_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272233216))), name = tensor("layers_11_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272172864))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_4604_cast_fp16 = conv(dilations = var_4604_dilations_0, groups = var_4604_groups_0, pad = var_4604_pad_0, pad_type = var_4604_pad_type_0, strides = var_4604_strides_0, weight = layers_11_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_311_cast_fp16)[name = tensor("op_4604_cast_fp16")]; + tensor input_313_cast_fp16 = add(x = var_4598_cast_fp16, y = var_4604_cast_fp16)[name = tensor("input_313_cast_fp16")]; + tensor input_315_split_num_splits_0 = const()[name = tensor("input_315_split_num_splits_0"), val = tensor(2)]; + tensor input_315_split_axis_0 = const()[name = tensor("input_315_split_axis_0"), val = tensor(1)]; + tensor input_315_split_cast_fp16_0, tensor input_315_split_cast_fp16_1 = split(axis = input_315_split_axis_0, num_splits = input_315_split_num_splits_0, x = input_313_cast_fp16)[name = tensor("input_315_split_cast_fp16")]; + tensor input_315_split_1_sigmoid_cast_fp16 = sigmoid(x = input_315_split_cast_fp16_1)[name = tensor("input_315_split_1_sigmoid_cast_fp16")]; + tensor input_315_cast_fp16 = mul(x = input_315_split_cast_fp16_0, y = input_315_split_1_sigmoid_cast_fp16)[name = tensor("input_315_cast_fp16")]; + tensor input_317_pad_type_0 = const()[name = tensor("input_317_pad_type_0"), val = tensor("custom")]; + tensor input_317_pad_0 = const()[name = tensor("input_317_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_317_groups_0 = const()[name = tensor("input_317_groups_0"), val = tensor(1024)]; + tensor input_317_strides_0 = const()[name = tensor("input_317_strides_0"), val = tensor([1, 1])]; + tensor input_317_dilations_0 = const()[name = tensor("input_317_dilations_0"), val = tensor([1, 1])]; + tensor const_290_to_fp16 = const()[name = tensor("const_290_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272495424)))]; + tensor const_291_to_fp16 = const()[name = tensor("const_291_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272513920)))]; + tensor input_319_cast_fp16 = conv(bias = const_291_to_fp16, dilations = input_317_dilations_0, groups = input_317_groups_0, pad = input_317_pad_0, pad_type = input_317_pad_type_0, strides = input_317_strides_0, weight = const_290_to_fp16, x = input_315_cast_fp16)[name = tensor("input_319_cast_fp16")]; + tensor input_321_cast_fp16 = silu(x = input_319_cast_fp16)[name = tensor("input_321_cast_fp16")]; + tensor var_4626_pad_type_0 = const()[name = tensor("op_4626_pad_type_0"), val = tensor("valid")]; + tensor var_4626_strides_0 = const()[name = tensor("op_4626_strides_0"), val = tensor([1, 1])]; + tensor var_4626_pad_0 = const()[name = tensor("op_4626_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4626_dilations_0 = const()[name = tensor("op_4626_dilations_0"), val = tensor([1, 1])]; + tensor var_4626_groups_0 = const()[name = tensor("op_4626_groups_0"), val = tensor(1)]; + tensor layers_11_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(272516032))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(273302528))), name = tensor("layers_11_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4626_cast_fp16 = conv(dilations = var_4626_dilations_0, groups = var_4626_groups_0, pad = var_4626_pad_0, pad_type = var_4626_pad_type_0, strides = var_4626_strides_0, weight = layers_11_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_321_cast_fp16)[name = tensor("op_4626_cast_fp16")]; + tensor var_4632_pad_type_0 = const()[name = tensor("op_4632_pad_type_0"), val = tensor("valid")]; + tensor var_4632_strides_0 = const()[name = tensor("op_4632_strides_0"), val = tensor([1, 1])]; + tensor var_4632_pad_0 = const()[name = tensor("op_4632_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4632_dilations_0 = const()[name = tensor("op_4632_dilations_0"), val = tensor([1, 1])]; + tensor var_4632_groups_0 = const()[name = tensor("op_4632_groups_0"), val = tensor(1)]; + tensor layers_11_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(273334592))), name = tensor("layers_11_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(273302720))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4632_cast_fp16 = conv(dilations = var_4632_dilations_0, groups = var_4632_groups_0, pad = var_4632_pad_0, pad_type = var_4632_pad_type_0, strides = var_4632_strides_0, weight = layers_11_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_321_cast_fp16)[name = tensor("op_4632_cast_fp16")]; + tensor x_71_cast_fp16 = add(x = var_4626_cast_fp16, y = var_4632_cast_fp16)[name = tensor("x_71_cast_fp16")]; + tensor inputs_117_cast_fp16 = add(x = inputs_115_cast_fp16, y = x_71_cast_fp16)[name = tensor("inputs_117_cast_fp16")]; + tensor out_117_axes_0 = const()[name = tensor("out_117_axes_0"), val = tensor([1])]; + tensor var_4643_to_fp16 = const()[name = tensor("op_4643_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_117_cast_fp16 = layer_norm(axes = out_117_axes_0, epsilon = var_4643_to_fp16, x = inputs_117_cast_fp16)[name = tensor("out_117_cast_fp16")]; + tensor input_323_gamma_0_to_fp16 = const()[name = tensor("input_323_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(273465728)))]; + tensor input_323_beta_0_to_fp16 = const()[name = tensor("input_323_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(273467840)))]; + tensor input_323_epsilon_0_to_fp16 = const()[name = tensor("input_323_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_323_cast_fp16 = batch_norm(beta = input_323_beta_0_to_fp16, epsilon = input_323_epsilon_0_to_fp16, gamma = input_323_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_117_cast_fp16)[name = tensor("input_323_cast_fp16")]; + tensor var_4663_pad_type_0 = const()[name = tensor("op_4663_pad_type_0"), val = tensor("valid")]; + tensor var_4663_strides_0 = const()[name = tensor("op_4663_strides_0"), val = tensor([1, 1])]; + tensor var_4663_pad_0 = const()[name = tensor("op_4663_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4663_dilations_0 = const()[name = tensor("op_4663_dilations_0"), val = tensor([1, 1])]; + tensor var_4663_groups_0 = const()[name = tensor("op_4663_groups_0"), val = tensor(1)]; + tensor layers_11_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(273469952))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(276615744))), name = tensor("layers_11_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_4663_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_4663_dilations_0, groups = var_4663_groups_0, pad = var_4663_pad_0, pad_type = var_4663_pad_type_0, strides = var_4663_strides_0, weight = layers_11_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_323_cast_fp16)[name = tensor("op_4663_cast_fp16")]; + tensor var_4669_pad_type_0 = const()[name = tensor("op_4669_pad_type_0"), val = tensor("valid")]; + tensor var_4669_strides_0 = const()[name = tensor("op_4669_strides_0"), val = tensor([1, 1])]; + tensor var_4669_pad_0 = const()[name = tensor("op_4669_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4669_dilations_0 = const()[name = tensor("op_4669_dilations_0"), val = tensor([1, 1])]; + tensor var_4669_groups_0 = const()[name = tensor("op_4669_groups_0"), val = tensor(1)]; + tensor layers_11_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(276778688))), name = tensor("layers_11_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(276615936))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_4669_cast_fp16 = conv(dilations = var_4669_dilations_0, groups = var_4669_groups_0, pad = var_4669_pad_0, pad_type = var_4669_pad_type_0, strides = var_4669_strides_0, weight = layers_11_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_323_cast_fp16)[name = tensor("op_4669_cast_fp16")]; + tensor input_325_cast_fp16 = add(x = var_4663_cast_fp16, y = var_4669_cast_fp16)[name = tensor("input_325_cast_fp16")]; + tensor input_327_cast_fp16 = silu(x = input_325_cast_fp16)[name = tensor("input_327_cast_fp16")]; + tensor var_4680_pad_type_0 = const()[name = tensor("op_4680_pad_type_0"), val = tensor("valid")]; + tensor var_4680_strides_0 = const()[name = tensor("op_4680_strides_0"), val = tensor([1, 1])]; + tensor var_4680_pad_0 = const()[name = tensor("op_4680_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4680_dilations_0 = const()[name = tensor("op_4680_dilations_0"), val = tensor([1, 1])]; + tensor var_4680_groups_0 = const()[name = tensor("op_4680_groups_0"), val = tensor(1)]; + tensor layers_11_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(277303040))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(280448832))), name = tensor("layers_11_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_4680_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4680_dilations_0, groups = var_4680_groups_0, pad = var_4680_pad_0, pad_type = var_4680_pad_type_0, strides = var_4680_strides_0, weight = layers_11_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_327_cast_fp16)[name = tensor("op_4680_cast_fp16")]; + tensor var_4686_pad_type_0 = const()[name = tensor("op_4686_pad_type_0"), val = tensor("valid")]; + tensor var_4686_strides_0 = const()[name = tensor("op_4686_strides_0"), val = tensor([1, 1])]; + tensor var_4686_pad_0 = const()[name = tensor("op_4686_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4686_dilations_0 = const()[name = tensor("op_4686_dilations_0"), val = tensor([1, 1])]; + tensor var_4686_groups_0 = const()[name = tensor("op_4686_groups_0"), val = tensor(1)]; + tensor layers_11_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(280615936))), name = tensor("layers_11_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(280449024))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_4686_cast_fp16 = conv(dilations = var_4686_dilations_0, groups = var_4686_groups_0, pad = var_4686_pad_0, pad_type = var_4686_pad_type_0, strides = var_4686_strides_0, weight = layers_11_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_327_cast_fp16)[name = tensor("op_4686_cast_fp16")]; + tensor x_73_cast_fp16 = add(x = var_4680_cast_fp16, y = var_4686_cast_fp16)[name = tensor("x_73_cast_fp16")]; + tensor var_4688_to_fp16 = const()[name = tensor("op_4688_to_fp16"), val = tensor(0x1p-1)]; + tensor var_4689_cast_fp16 = mul(x = x_73_cast_fp16, y = var_4688_to_fp16)[name = tensor("op_4689_cast_fp16")]; + tensor inputs_119_cast_fp16 = add(x = inputs_117_cast_fp16, y = var_4689_cast_fp16)[name = tensor("inputs_119_cast_fp16")]; + tensor out_119_axes_0 = const()[name = tensor("out_119_axes_0"), val = tensor([1])]; + tensor var_4699_to_fp16 = const()[name = tensor("op_4699_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_119_cast_fp16 = layer_norm(axes = out_119_axes_0, epsilon = var_4699_to_fp16, x = inputs_119_cast_fp16)[name = tensor("out_119_cast_fp16")]; + tensor inputs_121_gamma_0_to_fp16 = const()[name = tensor("inputs_121_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(281140288)))]; + tensor inputs_121_beta_0_to_fp16 = const()[name = tensor("inputs_121_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(281142400)))]; + tensor inputs_121_epsilon_0_to_fp16 = const()[name = tensor("inputs_121_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_121_cast_fp16 = batch_norm(beta = inputs_121_beta_0_to_fp16, epsilon = inputs_121_epsilon_0_to_fp16, gamma = inputs_121_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_119_cast_fp16)[name = tensor("inputs_121_cast_fp16")]; + tensor var_4713 = const()[name = tensor("op_4713"), val = tensor(3)]; + tensor out_121_axes_0 = const()[name = tensor("out_121_axes_0"), val = tensor([1])]; + tensor var_4744_to_fp16 = const()[name = tensor("op_4744_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_121_cast_fp16 = layer_norm(axes = out_121_axes_0, epsilon = var_4744_to_fp16, x = inputs_121_cast_fp16)[name = tensor("out_121_cast_fp16")]; + tensor input_329_gamma_0_to_fp16 = const()[name = tensor("input_329_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(281144512)))]; + tensor input_329_beta_0_to_fp16 = const()[name = tensor("input_329_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(281146624)))]; + tensor input_329_epsilon_0_to_fp16 = const()[name = tensor("input_329_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_329_cast_fp16 = batch_norm(beta = input_329_beta_0_to_fp16, epsilon = input_329_epsilon_0_to_fp16, gamma = input_329_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_121_cast_fp16)[name = tensor("input_329_cast_fp16")]; + tensor var_4764_pad_type_0 = const()[name = tensor("op_4764_pad_type_0"), val = tensor("valid")]; + tensor var_4764_strides_0 = const()[name = tensor("op_4764_strides_0"), val = tensor([1, 1])]; + tensor var_4764_pad_0 = const()[name = tensor("op_4764_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4764_dilations_0 = const()[name = tensor("op_4764_dilations_0"), val = tensor([1, 1])]; + tensor var_4764_groups_0 = const()[name = tensor("op_4764_groups_0"), val = tensor(1)]; + tensor layers_12_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(281148736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(284294528))), name = tensor("layers_12_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_4764_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_4764_dilations_0, groups = var_4764_groups_0, pad = var_4764_pad_0, pad_type = var_4764_pad_type_0, strides = var_4764_strides_0, weight = layers_12_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_329_cast_fp16)[name = tensor("op_4764_cast_fp16")]; + tensor var_4770_pad_type_0 = const()[name = tensor("op_4770_pad_type_0"), val = tensor("valid")]; + tensor var_4770_strides_0 = const()[name = tensor("op_4770_strides_0"), val = tensor([1, 1])]; + tensor var_4770_pad_0 = const()[name = tensor("op_4770_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4770_dilations_0 = const()[name = tensor("op_4770_dilations_0"), val = tensor([1, 1])]; + tensor var_4770_groups_0 = const()[name = tensor("op_4770_groups_0"), val = tensor(1)]; + tensor layers_12_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(284467776))), name = tensor("layers_12_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(284294720))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_4770_cast_fp16 = conv(dilations = var_4770_dilations_0, groups = var_4770_groups_0, pad = var_4770_pad_0, pad_type = var_4770_pad_type_0, strides = var_4770_strides_0, weight = layers_12_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_329_cast_fp16)[name = tensor("op_4770_cast_fp16")]; + tensor input_331_cast_fp16 = add(x = var_4764_cast_fp16, y = var_4770_cast_fp16)[name = tensor("input_331_cast_fp16")]; + tensor input_333_cast_fp16 = silu(x = input_331_cast_fp16)[name = tensor("input_333_cast_fp16")]; + tensor var_4781_pad_type_0 = const()[name = tensor("op_4781_pad_type_0"), val = tensor("valid")]; + tensor var_4781_strides_0 = const()[name = tensor("op_4781_strides_0"), val = tensor([1, 1])]; + tensor var_4781_pad_0 = const()[name = tensor("op_4781_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4781_dilations_0 = const()[name = tensor("op_4781_dilations_0"), val = tensor([1, 1])]; + tensor var_4781_groups_0 = const()[name = tensor("op_4781_groups_0"), val = tensor(1)]; + tensor layers_12_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(284992128))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(288137920))), name = tensor("layers_12_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_4781_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4781_dilations_0, groups = var_4781_groups_0, pad = var_4781_pad_0, pad_type = var_4781_pad_type_0, strides = var_4781_strides_0, weight = layers_12_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_333_cast_fp16)[name = tensor("op_4781_cast_fp16")]; + tensor var_4787_pad_type_0 = const()[name = tensor("op_4787_pad_type_0"), val = tensor("valid")]; + tensor var_4787_strides_0 = const()[name = tensor("op_4787_strides_0"), val = tensor([1, 1])]; + tensor var_4787_pad_0 = const()[name = tensor("op_4787_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4787_dilations_0 = const()[name = tensor("op_4787_dilations_0"), val = tensor([1, 1])]; + tensor var_4787_groups_0 = const()[name = tensor("op_4787_groups_0"), val = tensor(1)]; + tensor layers_12_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(288297920))), name = tensor("layers_12_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(288138112))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_4787_cast_fp16 = conv(dilations = var_4787_dilations_0, groups = var_4787_groups_0, pad = var_4787_pad_0, pad_type = var_4787_pad_type_0, strides = var_4787_strides_0, weight = layers_12_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_333_cast_fp16)[name = tensor("op_4787_cast_fp16")]; + tensor x_75_cast_fp16 = add(x = var_4781_cast_fp16, y = var_4787_cast_fp16)[name = tensor("x_75_cast_fp16")]; + tensor var_4789_to_fp16 = const()[name = tensor("op_4789_to_fp16"), val = tensor(0x1p-1)]; + tensor var_4790_cast_fp16 = mul(x = x_75_cast_fp16, y = var_4789_to_fp16)[name = tensor("op_4790_cast_fp16")]; + tensor inputs_123_cast_fp16 = add(x = inputs_121_cast_fp16, y = var_4790_cast_fp16)[name = tensor("inputs_123_cast_fp16")]; + tensor out_123_axes_0 = const()[name = tensor("out_123_axes_0"), val = tensor([1])]; + tensor var_4800_to_fp16 = const()[name = tensor("op_4800_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_123_cast_fp16 = layer_norm(axes = out_123_axes_0, epsilon = var_4800_to_fp16, x = inputs_123_cast_fp16)[name = tensor("out_123_cast_fp16")]; + tensor obj_51_gamma_0_to_fp16 = const()[name = tensor("obj_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(288822272)))]; + tensor obj_51_beta_0_to_fp16 = const()[name = tensor("obj_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(288824384)))]; + tensor obj_51_epsilon_0_to_fp16 = const()[name = tensor("obj_51_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_51_cast_fp16 = batch_norm(beta = obj_51_beta_0_to_fp16, epsilon = obj_51_epsilon_0_to_fp16, gamma = obj_51_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_123_cast_fp16)[name = tensor("obj_51_cast_fp16")]; + tensor var_4825_pad_type_0 = const()[name = tensor("op_4825_pad_type_0"), val = tensor("valid")]; + tensor var_4825_strides_0 = const()[name = tensor("op_4825_strides_0"), val = tensor([1, 1])]; + tensor var_4825_pad_0 = const()[name = tensor("op_4825_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4825_dilations_0 = const()[name = tensor("op_4825_dilations_0"), val = tensor([1, 1])]; + tensor var_4825_groups_0 = const()[name = tensor("op_4825_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(288826496))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289612992))), name = tensor("layers_12_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4825_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4825_dilations_0, groups = var_4825_groups_0, pad = var_4825_pad_0, pad_type = var_4825_pad_type_0, strides = var_4825_strides_0, weight = layers_12_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_51_cast_fp16)[name = tensor("op_4825_cast_fp16")]; + tensor var_4831_pad_type_0 = const()[name = tensor("op_4831_pad_type_0"), val = tensor("valid")]; + tensor var_4831_strides_0 = const()[name = tensor("op_4831_strides_0"), val = tensor([1, 1])]; + tensor var_4831_pad_0 = const()[name = tensor("op_4831_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4831_dilations_0 = const()[name = tensor("op_4831_dilations_0"), val = tensor([1, 1])]; + tensor var_4831_groups_0 = const()[name = tensor("op_4831_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289644352))), name = tensor("layers_12_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289613184))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4831_cast_fp16 = conv(dilations = var_4831_dilations_0, groups = var_4831_groups_0, pad = var_4831_pad_0, pad_type = var_4831_pad_type_0, strides = var_4831_strides_0, weight = layers_12_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_51_cast_fp16)[name = tensor("op_4831_cast_fp16")]; + tensor query_49_cast_fp16 = add(x = var_4825_cast_fp16, y = var_4831_cast_fp16)[name = tensor("query_49_cast_fp16")]; + tensor var_4840_pad_type_0 = const()[name = tensor("op_4840_pad_type_0"), val = tensor("valid")]; + tensor var_4840_strides_0 = const()[name = tensor("op_4840_strides_0"), val = tensor([1, 1])]; + tensor var_4840_pad_0 = const()[name = tensor("op_4840_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4840_dilations_0 = const()[name = tensor("op_4840_dilations_0"), val = tensor([1, 1])]; + tensor var_4840_groups_0 = const()[name = tensor("op_4840_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289775488))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290561984))), name = tensor("layers_12_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4840_cast_fp16 = conv(dilations = var_4840_dilations_0, groups = var_4840_groups_0, pad = var_4840_pad_0, pad_type = var_4840_pad_type_0, strides = var_4840_strides_0, weight = layers_12_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_51_cast_fp16)[name = tensor("op_4840_cast_fp16")]; + tensor var_4846_pad_type_0 = const()[name = tensor("op_4846_pad_type_0"), val = tensor("valid")]; + tensor var_4846_strides_0 = const()[name = tensor("op_4846_strides_0"), val = tensor([1, 1])]; + tensor var_4846_pad_0 = const()[name = tensor("op_4846_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4846_dilations_0 = const()[name = tensor("op_4846_dilations_0"), val = tensor([1, 1])]; + tensor var_4846_groups_0 = const()[name = tensor("op_4846_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290595136))), name = tensor("layers_12_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290562176))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4846_cast_fp16 = conv(dilations = var_4846_dilations_0, groups = var_4846_groups_0, pad = var_4846_pad_0, pad_type = var_4846_pad_type_0, strides = var_4846_strides_0, weight = layers_12_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_51_cast_fp16)[name = tensor("op_4846_cast_fp16")]; + tensor key_25_cast_fp16 = add(x = var_4840_cast_fp16, y = var_4846_cast_fp16)[name = tensor("key_25_cast_fp16")]; + tensor var_4856_pad_type_0 = const()[name = tensor("op_4856_pad_type_0"), val = tensor("valid")]; + tensor var_4856_strides_0 = const()[name = tensor("op_4856_strides_0"), val = tensor([1, 1])]; + tensor var_4856_pad_0 = const()[name = tensor("op_4856_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4856_dilations_0 = const()[name = tensor("op_4856_dilations_0"), val = tensor([1, 1])]; + tensor var_4856_groups_0 = const()[name = tensor("op_4856_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290726272))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291512768))), name = tensor("layers_12_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4856_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4856_dilations_0, groups = var_4856_groups_0, pad = var_4856_pad_0, pad_type = var_4856_pad_type_0, strides = var_4856_strides_0, weight = layers_12_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_51_cast_fp16)[name = tensor("op_4856_cast_fp16")]; + tensor var_4862_pad_type_0 = const()[name = tensor("op_4862_pad_type_0"), val = tensor("valid")]; + tensor var_4862_strides_0 = const()[name = tensor("op_4862_strides_0"), val = tensor([1, 1])]; + tensor var_4862_pad_0 = const()[name = tensor("op_4862_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4862_dilations_0 = const()[name = tensor("op_4862_dilations_0"), val = tensor([1, 1])]; + tensor var_4862_groups_0 = const()[name = tensor("op_4862_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291543744))), name = tensor("layers_12_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291512960))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4862_cast_fp16 = conv(dilations = var_4862_dilations_0, groups = var_4862_groups_0, pad = var_4862_pad_0, pad_type = var_4862_pad_type_0, strides = var_4862_strides_0, weight = layers_12_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_51_cast_fp16)[name = tensor("op_4862_cast_fp16")]; + tensor value_25_cast_fp16 = add(x = var_4856_cast_fp16, y = var_4862_cast_fp16)[name = tensor("value_25_cast_fp16")]; + tensor var_4865_to_fp16 = const()[name = tensor("op_4865_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291674880)))]; + tensor query_51_cast_fp16 = add(x = query_49_cast_fp16, y = var_4865_to_fp16)[name = tensor("query_51_cast_fp16")]; + tensor var_4868_to_fp16 = const()[name = tensor("op_4868_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291676992)))]; + tensor q_with_bias_v_25_cast_fp16 = add(x = query_49_cast_fp16, y = var_4868_to_fp16)[name = tensor("q_with_bias_v_25_cast_fp16")]; + tensor var_4878_pad_type_0 = const()[name = tensor("op_4878_pad_type_0"), val = tensor("valid")]; + tensor var_4878_strides_0 = const()[name = tensor("op_4878_strides_0"), val = tensor([1, 1])]; + tensor var_4878_pad_0 = const()[name = tensor("op_4878_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4878_dilations_0 = const()[name = tensor("op_4878_dilations_0"), val = tensor([1, 1])]; + tensor var_4878_groups_0 = const()[name = tensor("op_4878_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291679104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(292465600))), name = tensor("layers_12_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4878_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4878_dilations_0, groups = var_4878_groups_0, pad = var_4878_pad_0, pad_type = var_4878_pad_type_0, strides = var_4878_strides_0, weight = layers_12_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_4878_cast_fp16")]; + tensor var_4884_pad_type_0 = const()[name = tensor("op_4884_pad_type_0"), val = tensor("valid")]; + tensor var_4884_strides_0 = const()[name = tensor("op_4884_strides_0"), val = tensor([1, 1])]; + tensor var_4884_pad_0 = const()[name = tensor("op_4884_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4884_dilations_0 = const()[name = tensor("op_4884_dilations_0"), val = tensor([1, 1])]; + tensor var_4884_groups_0 = const()[name = tensor("op_4884_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(292533120))), name = tensor("layers_12_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(292465792))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4884_cast_fp16 = conv(dilations = var_4884_dilations_0, groups = var_4884_groups_0, pad = var_4884_pad_0, pad_type = var_4884_pad_type_0, strides = var_4884_strides_0, weight = layers_12_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_4884_cast_fp16")]; + tensor p_25_cast_fp16 = add(x = var_4878_cast_fp16, y = var_4884_cast_fp16)[name = tensor("p_25_cast_fp16")]; + tensor var_4888 = const()[name = tensor("op_4888"), val = tensor([1, 8, 128, 188])]; + tensor var_4889_cast_fp16 = reshape(shape = var_4888, x = q_with_bias_v_25_cast_fp16)[name = tensor("op_4889_cast_fp16")]; + tensor var_4890 = const()[name = tensor("op_4890"), val = tensor([1, 8, 128, -1])]; + tensor var_4891_cast_fp16 = reshape(shape = var_4890, x = p_25_cast_fp16)[name = tensor("op_4891_cast_fp16")]; + tensor matrix_bd_97_transpose_x_0 = const()[name = tensor("matrix_bd_97_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_97_transpose_y_0 = const()[name = tensor("matrix_bd_97_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_97_cast_fp16 = matmul(transpose_x = matrix_bd_97_transpose_x_0, transpose_y = matrix_bd_97_transpose_y_0, x = var_4889_cast_fp16, y = var_4891_cast_fp16)[name = tensor("matrix_bd_97_cast_fp16")]; + tensor matrix_bd_99_pad_0 = const()[name = tensor("matrix_bd_99_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_99_mode_0 = const()[name = tensor("matrix_bd_99_mode_0"), val = tensor("constant")]; + tensor const_142_to_fp16 = const()[name = tensor("const_142_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_99_cast_fp16 = pad(constant_val = const_142_to_fp16, mode = matrix_bd_99_mode_0, pad = matrix_bd_99_pad_0, x = matrix_bd_97_cast_fp16)[name = tensor("matrix_bd_99_cast_fp16")]; + tensor var_4900 = const()[name = tensor("op_4900"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_101_cast_fp16 = reshape(shape = var_4900, x = matrix_bd_99_cast_fp16)[name = tensor("matrix_bd_101_cast_fp16")]; + tensor var_4904_begin_0 = const()[name = tensor("op_4904_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_4904_end_0 = const()[name = tensor("op_4904_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_4904_end_mask_0 = const()[name = tensor("op_4904_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4904_cast_fp16 = slice_by_index(begin = var_4904_begin_0, end = var_4904_end_0, end_mask = var_4904_end_mask_0, x = matrix_bd_101_cast_fp16)[name = tensor("op_4904_cast_fp16")]; + tensor var_4905 = const()[name = tensor("op_4905"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_103_cast_fp16 = reshape(shape = var_4905, x = var_4904_cast_fp16)[name = tensor("matrix_bd_103_cast_fp16")]; + tensor var_4910_begin_0 = const()[name = tensor("op_4910_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4910_end_0 = const()[name = tensor("op_4910_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_4910_end_mask_0 = const()[name = tensor("op_4910_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4910_cast_fp16 = slice_by_index(begin = var_4910_begin_0, end = var_4910_end_0, end_mask = var_4910_end_mask_0, x = matrix_bd_103_cast_fp16)[name = tensor("op_4910_cast_fp16")]; + tensor var_4911_to_fp16 = const()[name = tensor("op_4911_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_25_cast_fp16 = mul(x = var_4910_cast_fp16, y = var_4911_to_fp16)[name = tensor("qk_mask_25_cast_fp16")]; + tensor var_4915 = const()[name = tensor("op_4915"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_25_cast_fp16 = reshape(shape = var_4915, x = query_51_cast_fp16)[name = tensor("mh_q_25_cast_fp16")]; + tensor var_4917_to_fp16 = const()[name = tensor("op_4917_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_4918_cast_fp16 = mul(x = mh_q_25_cast_fp16, y = var_4917_to_fp16)[name = tensor("op_4918_cast_fp16")]; + tensor var_4921 = const()[name = tensor("op_4921"), val = tensor([1, 8, 128, 188])]; + tensor var_4922_cast_fp16 = reshape(shape = var_4921, x = key_25_cast_fp16)[name = tensor("op_4922_cast_fp16")]; + tensor mh_w_49_transpose_x_0 = const()[name = tensor("mh_w_49_transpose_x_0"), val = tensor(true)]; + tensor mh_w_49_transpose_y_0 = const()[name = tensor("mh_w_49_transpose_y_0"), val = tensor(false)]; + tensor mh_w_49_cast_fp16 = matmul(transpose_x = mh_w_49_transpose_x_0, transpose_y = mh_w_49_transpose_y_0, x = var_4918_cast_fp16, y = var_4922_cast_fp16)[name = tensor("mh_w_49_cast_fp16")]; + tensor mh_w_51_cast_fp16 = add(x = mh_w_49_cast_fp16, y = qk_mask_25_cast_fp16)[name = tensor("mh_w_51_cast_fp16")]; + tensor var_4926_cast_fp16 = softmax(axis = var_4713, x = mh_w_51_cast_fp16)[name = tensor("op_4926_cast_fp16")]; + tensor var_4927 = const()[name = tensor("op_4927"), val = tensor([1, 8, 128, 188])]; + tensor var_4928_cast_fp16 = reshape(shape = var_4927, x = value_25_cast_fp16)[name = tensor("op_4928_cast_fp16")]; + tensor attn_25_transpose_x_0 = const()[name = tensor("attn_25_transpose_x_0"), val = tensor(false)]; + tensor attn_25_transpose_y_0 = const()[name = tensor("attn_25_transpose_y_0"), val = tensor(true)]; + tensor attn_25_cast_fp16 = matmul(transpose_x = attn_25_transpose_x_0, transpose_y = attn_25_transpose_y_0, x = var_4928_cast_fp16, y = var_4926_cast_fp16)[name = tensor("attn_25_cast_fp16")]; + tensor var_4931 = const()[name = tensor("op_4931"), val = tensor([1, 1024, 1, 188])]; + tensor input_335_cast_fp16 = reshape(shape = var_4931, x = attn_25_cast_fp16)[name = tensor("input_335_cast_fp16")]; + tensor var_4941_pad_type_0 = const()[name = tensor("op_4941_pad_type_0"), val = tensor("valid")]; + tensor var_4941_strides_0 = const()[name = tensor("op_4941_strides_0"), val = tensor([1, 1])]; + tensor var_4941_pad_0 = const()[name = tensor("op_4941_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4941_dilations_0 = const()[name = tensor("op_4941_dilations_0"), val = tensor([1, 1])]; + tensor var_4941_groups_0 = const()[name = tensor("op_4941_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(292664256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293450752))), name = tensor("layers_12_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4941_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4941_dilations_0, groups = var_4941_groups_0, pad = var_4941_pad_0, pad_type = var_4941_pad_type_0, strides = var_4941_strides_0, weight = layers_12_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_335_cast_fp16)[name = tensor("op_4941_cast_fp16")]; + tensor var_4947_pad_type_0 = const()[name = tensor("op_4947_pad_type_0"), val = tensor("valid")]; + tensor var_4947_strides_0 = const()[name = tensor("op_4947_strides_0"), val = tensor([1, 1])]; + tensor var_4947_pad_0 = const()[name = tensor("op_4947_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4947_dilations_0 = const()[name = tensor("op_4947_dilations_0"), val = tensor([1, 1])]; + tensor var_4947_groups_0 = const()[name = tensor("op_4947_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293484032))), name = tensor("layers_12_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293450944))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_4947_cast_fp16 = conv(dilations = var_4947_dilations_0, groups = var_4947_groups_0, pad = var_4947_pad_0, pad_type = var_4947_pad_type_0, strides = var_4947_strides_0, weight = layers_12_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_335_cast_fp16)[name = tensor("op_4947_cast_fp16")]; + tensor obj_53_cast_fp16 = add(x = var_4941_cast_fp16, y = var_4947_cast_fp16)[name = tensor("obj_53_cast_fp16")]; + tensor inputs_125_cast_fp16 = add(x = inputs_123_cast_fp16, y = obj_53_cast_fp16)[name = tensor("inputs_125_cast_fp16")]; + tensor out_125_axes_0 = const()[name = tensor("out_125_axes_0"), val = tensor([1])]; + tensor var_4958_to_fp16 = const()[name = tensor("op_4958_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_125_cast_fp16 = layer_norm(axes = out_125_axes_0, epsilon = var_4958_to_fp16, x = inputs_125_cast_fp16)[name = tensor("out_125_cast_fp16")]; + tensor input_337_gamma_0_to_fp16 = const()[name = tensor("input_337_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293615168)))]; + tensor input_337_beta_0_to_fp16 = const()[name = tensor("input_337_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293617280)))]; + tensor input_337_epsilon_0_to_fp16 = const()[name = tensor("input_337_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_337_cast_fp16 = batch_norm(beta = input_337_beta_0_to_fp16, epsilon = input_337_epsilon_0_to_fp16, gamma = input_337_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_125_cast_fp16)[name = tensor("input_337_cast_fp16")]; + tensor var_4979_pad_type_0 = const()[name = tensor("op_4979_pad_type_0"), val = tensor("valid")]; + tensor var_4979_strides_0 = const()[name = tensor("op_4979_strides_0"), val = tensor([1, 1])]; + tensor var_4979_pad_0 = const()[name = tensor("op_4979_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4979_dilations_0 = const()[name = tensor("op_4979_dilations_0"), val = tensor([1, 1])]; + tensor var_4979_groups_0 = const()[name = tensor("op_4979_groups_0"), val = tensor(1)]; + tensor layers_12_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293619392))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295192320))), name = tensor("layers_12_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_4979_cast_fp16 = conv(dilations = var_4979_dilations_0, groups = var_4979_groups_0, pad = var_4979_pad_0, pad_type = var_4979_pad_type_0, strides = var_4979_strides_0, weight = layers_12_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_337_cast_fp16)[name = tensor("op_4979_cast_fp16")]; + tensor var_4985_pad_type_0 = const()[name = tensor("op_4985_pad_type_0"), val = tensor("valid")]; + tensor var_4985_strides_0 = const()[name = tensor("op_4985_strides_0"), val = tensor([1, 1])]; + tensor var_4985_pad_0 = const()[name = tensor("op_4985_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4985_dilations_0 = const()[name = tensor("op_4985_dilations_0"), val = tensor([1, 1])]; + tensor var_4985_groups_0 = const()[name = tensor("op_4985_groups_0"), val = tensor(1)]; + tensor layers_12_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295256000))), name = tensor("layers_12_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295192512))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_4985_cast_fp16 = conv(dilations = var_4985_dilations_0, groups = var_4985_groups_0, pad = var_4985_pad_0, pad_type = var_4985_pad_type_0, strides = var_4985_strides_0, weight = layers_12_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_337_cast_fp16)[name = tensor("op_4985_cast_fp16")]; + tensor input_339_cast_fp16 = add(x = var_4979_cast_fp16, y = var_4985_cast_fp16)[name = tensor("input_339_cast_fp16")]; + tensor input_341_split_num_splits_0 = const()[name = tensor("input_341_split_num_splits_0"), val = tensor(2)]; + tensor input_341_split_axis_0 = const()[name = tensor("input_341_split_axis_0"), val = tensor(1)]; + tensor input_341_split_cast_fp16_0, tensor input_341_split_cast_fp16_1 = split(axis = input_341_split_axis_0, num_splits = input_341_split_num_splits_0, x = input_339_cast_fp16)[name = tensor("input_341_split_cast_fp16")]; + tensor input_341_split_1_sigmoid_cast_fp16 = sigmoid(x = input_341_split_cast_fp16_1)[name = tensor("input_341_split_1_sigmoid_cast_fp16")]; + tensor input_341_cast_fp16 = mul(x = input_341_split_cast_fp16_0, y = input_341_split_1_sigmoid_cast_fp16)[name = tensor("input_341_cast_fp16")]; + tensor input_343_pad_type_0 = const()[name = tensor("input_343_pad_type_0"), val = tensor("custom")]; + tensor input_343_pad_0 = const()[name = tensor("input_343_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_343_groups_0 = const()[name = tensor("input_343_groups_0"), val = tensor(1024)]; + tensor input_343_strides_0 = const()[name = tensor("input_343_strides_0"), val = tensor([1, 1])]; + tensor input_343_dilations_0 = const()[name = tensor("input_343_dilations_0"), val = tensor([1, 1])]; + tensor const_292_to_fp16 = const()[name = tensor("const_292_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295518208)))]; + tensor const_293_to_fp16 = const()[name = tensor("const_293_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295536704)))]; + tensor input_345_cast_fp16 = conv(bias = const_293_to_fp16, dilations = input_343_dilations_0, groups = input_343_groups_0, pad = input_343_pad_0, pad_type = input_343_pad_type_0, strides = input_343_strides_0, weight = const_292_to_fp16, x = input_341_cast_fp16)[name = tensor("input_345_cast_fp16")]; + tensor input_347_cast_fp16 = silu(x = input_345_cast_fp16)[name = tensor("input_347_cast_fp16")]; + tensor var_5007_pad_type_0 = const()[name = tensor("op_5007_pad_type_0"), val = tensor("valid")]; + tensor var_5007_strides_0 = const()[name = tensor("op_5007_strides_0"), val = tensor([1, 1])]; + tensor var_5007_pad_0 = const()[name = tensor("op_5007_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5007_dilations_0 = const()[name = tensor("op_5007_dilations_0"), val = tensor([1, 1])]; + tensor var_5007_groups_0 = const()[name = tensor("op_5007_groups_0"), val = tensor(1)]; + tensor layers_12_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295538816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296325312))), name = tensor("layers_12_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5007_cast_fp16 = conv(dilations = var_5007_dilations_0, groups = var_5007_groups_0, pad = var_5007_pad_0, pad_type = var_5007_pad_type_0, strides = var_5007_strides_0, weight = layers_12_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_347_cast_fp16)[name = tensor("op_5007_cast_fp16")]; + tensor var_5013_pad_type_0 = const()[name = tensor("op_5013_pad_type_0"), val = tensor("valid")]; + tensor var_5013_strides_0 = const()[name = tensor("op_5013_strides_0"), val = tensor([1, 1])]; + tensor var_5013_pad_0 = const()[name = tensor("op_5013_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5013_dilations_0 = const()[name = tensor("op_5013_dilations_0"), val = tensor([1, 1])]; + tensor var_5013_groups_0 = const()[name = tensor("op_5013_groups_0"), val = tensor(1)]; + tensor layers_12_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296357440))), name = tensor("layers_12_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296325504))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5013_cast_fp16 = conv(dilations = var_5013_dilations_0, groups = var_5013_groups_0, pad = var_5013_pad_0, pad_type = var_5013_pad_type_0, strides = var_5013_strides_0, weight = layers_12_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_347_cast_fp16)[name = tensor("op_5013_cast_fp16")]; + tensor x_77_cast_fp16 = add(x = var_5007_cast_fp16, y = var_5013_cast_fp16)[name = tensor("x_77_cast_fp16")]; + tensor inputs_127_cast_fp16 = add(x = inputs_125_cast_fp16, y = x_77_cast_fp16)[name = tensor("inputs_127_cast_fp16")]; + tensor out_127_axes_0 = const()[name = tensor("out_127_axes_0"), val = tensor([1])]; + tensor var_5024_to_fp16 = const()[name = tensor("op_5024_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_127_cast_fp16 = layer_norm(axes = out_127_axes_0, epsilon = var_5024_to_fp16, x = inputs_127_cast_fp16)[name = tensor("out_127_cast_fp16")]; + tensor input_349_gamma_0_to_fp16 = const()[name = tensor("input_349_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296488576)))]; + tensor input_349_beta_0_to_fp16 = const()[name = tensor("input_349_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296490688)))]; + tensor input_349_epsilon_0_to_fp16 = const()[name = tensor("input_349_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_349_cast_fp16 = batch_norm(beta = input_349_beta_0_to_fp16, epsilon = input_349_epsilon_0_to_fp16, gamma = input_349_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_127_cast_fp16)[name = tensor("input_349_cast_fp16")]; + tensor var_5044_pad_type_0 = const()[name = tensor("op_5044_pad_type_0"), val = tensor("valid")]; + tensor var_5044_strides_0 = const()[name = tensor("op_5044_strides_0"), val = tensor([1, 1])]; + tensor var_5044_pad_0 = const()[name = tensor("op_5044_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5044_dilations_0 = const()[name = tensor("op_5044_dilations_0"), val = tensor([1, 1])]; + tensor var_5044_groups_0 = const()[name = tensor("op_5044_groups_0"), val = tensor(1)]; + tensor layers_12_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296492800))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(299638592))), name = tensor("layers_12_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_5044_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_5044_dilations_0, groups = var_5044_groups_0, pad = var_5044_pad_0, pad_type = var_5044_pad_type_0, strides = var_5044_strides_0, weight = layers_12_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_349_cast_fp16)[name = tensor("op_5044_cast_fp16")]; + tensor var_5050_pad_type_0 = const()[name = tensor("op_5050_pad_type_0"), val = tensor("valid")]; + tensor var_5050_strides_0 = const()[name = tensor("op_5050_strides_0"), val = tensor([1, 1])]; + tensor var_5050_pad_0 = const()[name = tensor("op_5050_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5050_dilations_0 = const()[name = tensor("op_5050_dilations_0"), val = tensor([1, 1])]; + tensor var_5050_groups_0 = const()[name = tensor("op_5050_groups_0"), val = tensor(1)]; + tensor layers_12_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(299797632))), name = tensor("layers_12_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(299638784))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_5050_cast_fp16 = conv(dilations = var_5050_dilations_0, groups = var_5050_groups_0, pad = var_5050_pad_0, pad_type = var_5050_pad_type_0, strides = var_5050_strides_0, weight = layers_12_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_349_cast_fp16)[name = tensor("op_5050_cast_fp16")]; + tensor input_351_cast_fp16 = add(x = var_5044_cast_fp16, y = var_5050_cast_fp16)[name = tensor("input_351_cast_fp16")]; + tensor input_353_cast_fp16 = silu(x = input_351_cast_fp16)[name = tensor("input_353_cast_fp16")]; + tensor var_5061_pad_type_0 = const()[name = tensor("op_5061_pad_type_0"), val = tensor("valid")]; + tensor var_5061_strides_0 = const()[name = tensor("op_5061_strides_0"), val = tensor([1, 1])]; + tensor var_5061_pad_0 = const()[name = tensor("op_5061_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5061_dilations_0 = const()[name = tensor("op_5061_dilations_0"), val = tensor([1, 1])]; + tensor var_5061_groups_0 = const()[name = tensor("op_5061_groups_0"), val = tensor(1)]; + tensor layers_12_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(300321984))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303467776))), name = tensor("layers_12_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_5061_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5061_dilations_0, groups = var_5061_groups_0, pad = var_5061_pad_0, pad_type = var_5061_pad_type_0, strides = var_5061_strides_0, weight = layers_12_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_353_cast_fp16)[name = tensor("op_5061_cast_fp16")]; + tensor var_5067_pad_type_0 = const()[name = tensor("op_5067_pad_type_0"), val = tensor("valid")]; + tensor var_5067_strides_0 = const()[name = tensor("op_5067_strides_0"), val = tensor([1, 1])]; + tensor var_5067_pad_0 = const()[name = tensor("op_5067_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5067_dilations_0 = const()[name = tensor("op_5067_dilations_0"), val = tensor([1, 1])]; + tensor var_5067_groups_0 = const()[name = tensor("op_5067_groups_0"), val = tensor(1)]; + tensor layers_12_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303634944))), name = tensor("layers_12_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303467968))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_5067_cast_fp16 = conv(dilations = var_5067_dilations_0, groups = var_5067_groups_0, pad = var_5067_pad_0, pad_type = var_5067_pad_type_0, strides = var_5067_strides_0, weight = layers_12_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_353_cast_fp16)[name = tensor("op_5067_cast_fp16")]; + tensor x_79_cast_fp16 = add(x = var_5061_cast_fp16, y = var_5067_cast_fp16)[name = tensor("x_79_cast_fp16")]; + tensor var_5069_to_fp16 = const()[name = tensor("op_5069_to_fp16"), val = tensor(0x1p-1)]; + tensor var_5070_cast_fp16 = mul(x = x_79_cast_fp16, y = var_5069_to_fp16)[name = tensor("op_5070_cast_fp16")]; + tensor inputs_129_cast_fp16 = add(x = inputs_127_cast_fp16, y = var_5070_cast_fp16)[name = tensor("inputs_129_cast_fp16")]; + tensor out_129_axes_0 = const()[name = tensor("out_129_axes_0"), val = tensor([1])]; + tensor var_5080_to_fp16 = const()[name = tensor("op_5080_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_129_cast_fp16 = layer_norm(axes = out_129_axes_0, epsilon = var_5080_to_fp16, x = inputs_129_cast_fp16)[name = tensor("out_129_cast_fp16")]; + tensor inputs_131_gamma_0_to_fp16 = const()[name = tensor("inputs_131_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304159296)))]; + tensor inputs_131_beta_0_to_fp16 = const()[name = tensor("inputs_131_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304161408)))]; + tensor inputs_131_epsilon_0_to_fp16 = const()[name = tensor("inputs_131_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_131_cast_fp16 = batch_norm(beta = inputs_131_beta_0_to_fp16, epsilon = inputs_131_epsilon_0_to_fp16, gamma = inputs_131_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_129_cast_fp16)[name = tensor("inputs_131_cast_fp16")]; + tensor var_5094 = const()[name = tensor("op_5094"), val = tensor(3)]; + tensor out_131_axes_0 = const()[name = tensor("out_131_axes_0"), val = tensor([1])]; + tensor var_5125_to_fp16 = const()[name = tensor("op_5125_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_131_cast_fp16 = layer_norm(axes = out_131_axes_0, epsilon = var_5125_to_fp16, x = inputs_131_cast_fp16)[name = tensor("out_131_cast_fp16")]; + tensor input_355_gamma_0_to_fp16 = const()[name = tensor("input_355_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304163520)))]; + tensor input_355_beta_0_to_fp16 = const()[name = tensor("input_355_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304165632)))]; + tensor input_355_epsilon_0_to_fp16 = const()[name = tensor("input_355_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_355_cast_fp16 = batch_norm(beta = input_355_beta_0_to_fp16, epsilon = input_355_epsilon_0_to_fp16, gamma = input_355_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_131_cast_fp16)[name = tensor("input_355_cast_fp16")]; + tensor var_5145_pad_type_0 = const()[name = tensor("op_5145_pad_type_0"), val = tensor("valid")]; + tensor var_5145_strides_0 = const()[name = tensor("op_5145_strides_0"), val = tensor([1, 1])]; + tensor var_5145_pad_0 = const()[name = tensor("op_5145_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5145_dilations_0 = const()[name = tensor("op_5145_dilations_0"), val = tensor([1, 1])]; + tensor var_5145_groups_0 = const()[name = tensor("op_5145_groups_0"), val = tensor(1)]; + tensor layers_13_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(304167744))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(307313536))), name = tensor("layers_13_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_5145_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_5145_dilations_0, groups = var_5145_groups_0, pad = var_5145_pad_0, pad_type = var_5145_pad_type_0, strides = var_5145_strides_0, weight = layers_13_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_355_cast_fp16)[name = tensor("op_5145_cast_fp16")]; + tensor var_5151_pad_type_0 = const()[name = tensor("op_5151_pad_type_0"), val = tensor("valid")]; + tensor var_5151_strides_0 = const()[name = tensor("op_5151_strides_0"), val = tensor([1, 1])]; + tensor var_5151_pad_0 = const()[name = tensor("op_5151_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5151_dilations_0 = const()[name = tensor("op_5151_dilations_0"), val = tensor([1, 1])]; + tensor var_5151_groups_0 = const()[name = tensor("op_5151_groups_0"), val = tensor(1)]; + tensor layers_13_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(307492672))), name = tensor("layers_13_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(307313728))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_5151_cast_fp16 = conv(dilations = var_5151_dilations_0, groups = var_5151_groups_0, pad = var_5151_pad_0, pad_type = var_5151_pad_type_0, strides = var_5151_strides_0, weight = layers_13_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_355_cast_fp16)[name = tensor("op_5151_cast_fp16")]; + tensor input_357_cast_fp16 = add(x = var_5145_cast_fp16, y = var_5151_cast_fp16)[name = tensor("input_357_cast_fp16")]; + tensor input_359_cast_fp16 = silu(x = input_357_cast_fp16)[name = tensor("input_359_cast_fp16")]; + tensor var_5162_pad_type_0 = const()[name = tensor("op_5162_pad_type_0"), val = tensor("valid")]; + tensor var_5162_strides_0 = const()[name = tensor("op_5162_strides_0"), val = tensor([1, 1])]; + tensor var_5162_pad_0 = const()[name = tensor("op_5162_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5162_dilations_0 = const()[name = tensor("op_5162_dilations_0"), val = tensor([1, 1])]; + tensor var_5162_groups_0 = const()[name = tensor("op_5162_groups_0"), val = tensor(1)]; + tensor layers_13_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(308017024))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311162816))), name = tensor("layers_13_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_5162_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5162_dilations_0, groups = var_5162_groups_0, pad = var_5162_pad_0, pad_type = var_5162_pad_type_0, strides = var_5162_strides_0, weight = layers_13_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_359_cast_fp16)[name = tensor("op_5162_cast_fp16")]; + tensor var_5168_pad_type_0 = const()[name = tensor("op_5168_pad_type_0"), val = tensor("valid")]; + tensor var_5168_strides_0 = const()[name = tensor("op_5168_strides_0"), val = tensor([1, 1])]; + tensor var_5168_pad_0 = const()[name = tensor("op_5168_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5168_dilations_0 = const()[name = tensor("op_5168_dilations_0"), val = tensor([1, 1])]; + tensor var_5168_groups_0 = const()[name = tensor("op_5168_groups_0"), val = tensor(1)]; + tensor layers_13_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311334848))), name = tensor("layers_13_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311163008))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_5168_cast_fp16 = conv(dilations = var_5168_dilations_0, groups = var_5168_groups_0, pad = var_5168_pad_0, pad_type = var_5168_pad_type_0, strides = var_5168_strides_0, weight = layers_13_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_359_cast_fp16)[name = tensor("op_5168_cast_fp16")]; + tensor x_81_cast_fp16 = add(x = var_5162_cast_fp16, y = var_5168_cast_fp16)[name = tensor("x_81_cast_fp16")]; + tensor var_5170_to_fp16 = const()[name = tensor("op_5170_to_fp16"), val = tensor(0x1p-1)]; + tensor var_5171_cast_fp16 = mul(x = x_81_cast_fp16, y = var_5170_to_fp16)[name = tensor("op_5171_cast_fp16")]; + tensor inputs_133_cast_fp16 = add(x = inputs_131_cast_fp16, y = var_5171_cast_fp16)[name = tensor("inputs_133_cast_fp16")]; + tensor out_133_axes_0 = const()[name = tensor("out_133_axes_0"), val = tensor([1])]; + tensor var_5181_to_fp16 = const()[name = tensor("op_5181_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_133_cast_fp16 = layer_norm(axes = out_133_axes_0, epsilon = var_5181_to_fp16, x = inputs_133_cast_fp16)[name = tensor("out_133_cast_fp16")]; + tensor obj_55_gamma_0_to_fp16 = const()[name = tensor("obj_55_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311859200)))]; + tensor obj_55_beta_0_to_fp16 = const()[name = tensor("obj_55_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311861312)))]; + tensor obj_55_epsilon_0_to_fp16 = const()[name = tensor("obj_55_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_55_cast_fp16 = batch_norm(beta = obj_55_beta_0_to_fp16, epsilon = obj_55_epsilon_0_to_fp16, gamma = obj_55_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_133_cast_fp16)[name = tensor("obj_55_cast_fp16")]; + tensor var_5206_pad_type_0 = const()[name = tensor("op_5206_pad_type_0"), val = tensor("valid")]; + tensor var_5206_strides_0 = const()[name = tensor("op_5206_strides_0"), val = tensor([1, 1])]; + tensor var_5206_pad_0 = const()[name = tensor("op_5206_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5206_dilations_0 = const()[name = tensor("op_5206_dilations_0"), val = tensor([1, 1])]; + tensor var_5206_groups_0 = const()[name = tensor("op_5206_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(311863424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312649920))), name = tensor("layers_13_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5206_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5206_dilations_0, groups = var_5206_groups_0, pad = var_5206_pad_0, pad_type = var_5206_pad_type_0, strides = var_5206_strides_0, weight = layers_13_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_55_cast_fp16)[name = tensor("op_5206_cast_fp16")]; + tensor var_5212_pad_type_0 = const()[name = tensor("op_5212_pad_type_0"), val = tensor("valid")]; + tensor var_5212_strides_0 = const()[name = tensor("op_5212_strides_0"), val = tensor([1, 1])]; + tensor var_5212_pad_0 = const()[name = tensor("op_5212_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5212_dilations_0 = const()[name = tensor("op_5212_dilations_0"), val = tensor([1, 1])]; + tensor var_5212_groups_0 = const()[name = tensor("op_5212_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312685504))), name = tensor("layers_13_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312650112))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5212_cast_fp16 = conv(dilations = var_5212_dilations_0, groups = var_5212_groups_0, pad = var_5212_pad_0, pad_type = var_5212_pad_type_0, strides = var_5212_strides_0, weight = layers_13_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_55_cast_fp16)[name = tensor("op_5212_cast_fp16")]; + tensor query_53_cast_fp16 = add(x = var_5206_cast_fp16, y = var_5212_cast_fp16)[name = tensor("query_53_cast_fp16")]; + tensor var_5221_pad_type_0 = const()[name = tensor("op_5221_pad_type_0"), val = tensor("valid")]; + tensor var_5221_strides_0 = const()[name = tensor("op_5221_strides_0"), val = tensor([1, 1])]; + tensor var_5221_pad_0 = const()[name = tensor("op_5221_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5221_dilations_0 = const()[name = tensor("op_5221_dilations_0"), val = tensor([1, 1])]; + tensor var_5221_groups_0 = const()[name = tensor("op_5221_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312816640))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(313603136))), name = tensor("layers_13_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5221_cast_fp16 = conv(dilations = var_5221_dilations_0, groups = var_5221_groups_0, pad = var_5221_pad_0, pad_type = var_5221_pad_type_0, strides = var_5221_strides_0, weight = layers_13_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_55_cast_fp16)[name = tensor("op_5221_cast_fp16")]; + tensor var_5227_pad_type_0 = const()[name = tensor("op_5227_pad_type_0"), val = tensor("valid")]; + tensor var_5227_strides_0 = const()[name = tensor("op_5227_strides_0"), val = tensor([1, 1])]; + tensor var_5227_pad_0 = const()[name = tensor("op_5227_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5227_dilations_0 = const()[name = tensor("op_5227_dilations_0"), val = tensor([1, 1])]; + tensor var_5227_groups_0 = const()[name = tensor("op_5227_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(313642752))), name = tensor("layers_13_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(313603328))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5227_cast_fp16 = conv(dilations = var_5227_dilations_0, groups = var_5227_groups_0, pad = var_5227_pad_0, pad_type = var_5227_pad_type_0, strides = var_5227_strides_0, weight = layers_13_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_55_cast_fp16)[name = tensor("op_5227_cast_fp16")]; + tensor key_27_cast_fp16 = add(x = var_5221_cast_fp16, y = var_5227_cast_fp16)[name = tensor("key_27_cast_fp16")]; + tensor var_5237_pad_type_0 = const()[name = tensor("op_5237_pad_type_0"), val = tensor("valid")]; + tensor var_5237_strides_0 = const()[name = tensor("op_5237_strides_0"), val = tensor([1, 1])]; + tensor var_5237_pad_0 = const()[name = tensor("op_5237_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5237_dilations_0 = const()[name = tensor("op_5237_dilations_0"), val = tensor([1, 1])]; + tensor var_5237_groups_0 = const()[name = tensor("op_5237_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(313773888))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314560384))), name = tensor("layers_13_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5237_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5237_dilations_0, groups = var_5237_groups_0, pad = var_5237_pad_0, pad_type = var_5237_pad_type_0, strides = var_5237_strides_0, weight = layers_13_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_55_cast_fp16)[name = tensor("op_5237_cast_fp16")]; + tensor var_5243_pad_type_0 = const()[name = tensor("op_5243_pad_type_0"), val = tensor("valid")]; + tensor var_5243_strides_0 = const()[name = tensor("op_5243_strides_0"), val = tensor([1, 1])]; + tensor var_5243_pad_0 = const()[name = tensor("op_5243_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5243_dilations_0 = const()[name = tensor("op_5243_dilations_0"), val = tensor([1, 1])]; + tensor var_5243_groups_0 = const()[name = tensor("op_5243_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314591744))), name = tensor("layers_13_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314560576))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5243_cast_fp16 = conv(dilations = var_5243_dilations_0, groups = var_5243_groups_0, pad = var_5243_pad_0, pad_type = var_5243_pad_type_0, strides = var_5243_strides_0, weight = layers_13_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_55_cast_fp16)[name = tensor("op_5243_cast_fp16")]; + tensor value_27_cast_fp16 = add(x = var_5237_cast_fp16, y = var_5243_cast_fp16)[name = tensor("value_27_cast_fp16")]; + tensor var_5246_to_fp16 = const()[name = tensor("op_5246_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314722880)))]; + tensor query_55_cast_fp16 = add(x = query_53_cast_fp16, y = var_5246_to_fp16)[name = tensor("query_55_cast_fp16")]; + tensor var_5249_to_fp16 = const()[name = tensor("op_5249_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314724992)))]; + tensor q_with_bias_v_27_cast_fp16 = add(x = query_53_cast_fp16, y = var_5249_to_fp16)[name = tensor("q_with_bias_v_27_cast_fp16")]; + tensor var_5259_pad_type_0 = const()[name = tensor("op_5259_pad_type_0"), val = tensor("valid")]; + tensor var_5259_strides_0 = const()[name = tensor("op_5259_strides_0"), val = tensor([1, 1])]; + tensor var_5259_pad_0 = const()[name = tensor("op_5259_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5259_dilations_0 = const()[name = tensor("op_5259_dilations_0"), val = tensor([1, 1])]; + tensor var_5259_groups_0 = const()[name = tensor("op_5259_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314727104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(315513600))), name = tensor("layers_13_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5259_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5259_dilations_0, groups = var_5259_groups_0, pad = var_5259_pad_0, pad_type = var_5259_pad_type_0, strides = var_5259_strides_0, weight = layers_13_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_5259_cast_fp16")]; + tensor var_5265_pad_type_0 = const()[name = tensor("op_5265_pad_type_0"), val = tensor("valid")]; + tensor var_5265_strides_0 = const()[name = tensor("op_5265_strides_0"), val = tensor([1, 1])]; + tensor var_5265_pad_0 = const()[name = tensor("op_5265_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5265_dilations_0 = const()[name = tensor("op_5265_dilations_0"), val = tensor([1, 1])]; + tensor var_5265_groups_0 = const()[name = tensor("op_5265_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(315581440))), name = tensor("layers_13_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(315513792))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5265_cast_fp16 = conv(dilations = var_5265_dilations_0, groups = var_5265_groups_0, pad = var_5265_pad_0, pad_type = var_5265_pad_type_0, strides = var_5265_strides_0, weight = layers_13_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_5265_cast_fp16")]; + tensor p_27_cast_fp16 = add(x = var_5259_cast_fp16, y = var_5265_cast_fp16)[name = tensor("p_27_cast_fp16")]; + tensor var_5269 = const()[name = tensor("op_5269"), val = tensor([1, 8, 128, 188])]; + tensor var_5270_cast_fp16 = reshape(shape = var_5269, x = q_with_bias_v_27_cast_fp16)[name = tensor("op_5270_cast_fp16")]; + tensor var_5271 = const()[name = tensor("op_5271"), val = tensor([1, 8, 128, -1])]; + tensor var_5272_cast_fp16 = reshape(shape = var_5271, x = p_27_cast_fp16)[name = tensor("op_5272_cast_fp16")]; + tensor matrix_bd_105_transpose_x_0 = const()[name = tensor("matrix_bd_105_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_105_transpose_y_0 = const()[name = tensor("matrix_bd_105_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_105_cast_fp16 = matmul(transpose_x = matrix_bd_105_transpose_x_0, transpose_y = matrix_bd_105_transpose_y_0, x = var_5270_cast_fp16, y = var_5272_cast_fp16)[name = tensor("matrix_bd_105_cast_fp16")]; + tensor matrix_bd_107_pad_0 = const()[name = tensor("matrix_bd_107_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_107_mode_0 = const()[name = tensor("matrix_bd_107_mode_0"), val = tensor("constant")]; + tensor const_153_to_fp16 = const()[name = tensor("const_153_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_107_cast_fp16 = pad(constant_val = const_153_to_fp16, mode = matrix_bd_107_mode_0, pad = matrix_bd_107_pad_0, x = matrix_bd_105_cast_fp16)[name = tensor("matrix_bd_107_cast_fp16")]; + tensor var_5281 = const()[name = tensor("op_5281"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_109_cast_fp16 = reshape(shape = var_5281, x = matrix_bd_107_cast_fp16)[name = tensor("matrix_bd_109_cast_fp16")]; + tensor var_5285_begin_0 = const()[name = tensor("op_5285_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_5285_end_0 = const()[name = tensor("op_5285_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_5285_end_mask_0 = const()[name = tensor("op_5285_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_5285_cast_fp16 = slice_by_index(begin = var_5285_begin_0, end = var_5285_end_0, end_mask = var_5285_end_mask_0, x = matrix_bd_109_cast_fp16)[name = tensor("op_5285_cast_fp16")]; + tensor var_5286 = const()[name = tensor("op_5286"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_111_cast_fp16 = reshape(shape = var_5286, x = var_5285_cast_fp16)[name = tensor("matrix_bd_111_cast_fp16")]; + tensor var_5291_begin_0 = const()[name = tensor("op_5291_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5291_end_0 = const()[name = tensor("op_5291_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_5291_end_mask_0 = const()[name = tensor("op_5291_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5291_cast_fp16 = slice_by_index(begin = var_5291_begin_0, end = var_5291_end_0, end_mask = var_5291_end_mask_0, x = matrix_bd_111_cast_fp16)[name = tensor("op_5291_cast_fp16")]; + tensor var_5292_to_fp16 = const()[name = tensor("op_5292_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_27_cast_fp16 = mul(x = var_5291_cast_fp16, y = var_5292_to_fp16)[name = tensor("qk_mask_27_cast_fp16")]; + tensor var_5296 = const()[name = tensor("op_5296"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_27_cast_fp16 = reshape(shape = var_5296, x = query_55_cast_fp16)[name = tensor("mh_q_27_cast_fp16")]; + tensor var_5298_to_fp16 = const()[name = tensor("op_5298_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_5299_cast_fp16 = mul(x = mh_q_27_cast_fp16, y = var_5298_to_fp16)[name = tensor("op_5299_cast_fp16")]; + tensor var_5302 = const()[name = tensor("op_5302"), val = tensor([1, 8, 128, 188])]; + tensor var_5303_cast_fp16 = reshape(shape = var_5302, x = key_27_cast_fp16)[name = tensor("op_5303_cast_fp16")]; + tensor mh_w_53_transpose_x_0 = const()[name = tensor("mh_w_53_transpose_x_0"), val = tensor(true)]; + tensor mh_w_53_transpose_y_0 = const()[name = tensor("mh_w_53_transpose_y_0"), val = tensor(false)]; + tensor mh_w_53_cast_fp16 = matmul(transpose_x = mh_w_53_transpose_x_0, transpose_y = mh_w_53_transpose_y_0, x = var_5299_cast_fp16, y = var_5303_cast_fp16)[name = tensor("mh_w_53_cast_fp16")]; + tensor mh_w_55_cast_fp16 = add(x = mh_w_53_cast_fp16, y = qk_mask_27_cast_fp16)[name = tensor("mh_w_55_cast_fp16")]; + tensor var_5307_cast_fp16 = softmax(axis = var_5094, x = mh_w_55_cast_fp16)[name = tensor("op_5307_cast_fp16")]; + tensor var_5308 = const()[name = tensor("op_5308"), val = tensor([1, 8, 128, 188])]; + tensor var_5309_cast_fp16 = reshape(shape = var_5308, x = value_27_cast_fp16)[name = tensor("op_5309_cast_fp16")]; + tensor attn_27_transpose_x_0 = const()[name = tensor("attn_27_transpose_x_0"), val = tensor(false)]; + tensor attn_27_transpose_y_0 = const()[name = tensor("attn_27_transpose_y_0"), val = tensor(true)]; + tensor attn_27_cast_fp16 = matmul(transpose_x = attn_27_transpose_x_0, transpose_y = attn_27_transpose_y_0, x = var_5309_cast_fp16, y = var_5307_cast_fp16)[name = tensor("attn_27_cast_fp16")]; + tensor var_5312 = const()[name = tensor("op_5312"), val = tensor([1, 1024, 1, 188])]; + tensor input_361_cast_fp16 = reshape(shape = var_5312, x = attn_27_cast_fp16)[name = tensor("input_361_cast_fp16")]; + tensor var_5322_pad_type_0 = const()[name = tensor("op_5322_pad_type_0"), val = tensor("valid")]; + tensor var_5322_strides_0 = const()[name = tensor("op_5322_strides_0"), val = tensor([1, 1])]; + tensor var_5322_pad_0 = const()[name = tensor("op_5322_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5322_dilations_0 = const()[name = tensor("op_5322_dilations_0"), val = tensor([1, 1])]; + tensor var_5322_groups_0 = const()[name = tensor("op_5322_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(315712576))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316499072))), name = tensor("layers_13_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5322_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5322_dilations_0, groups = var_5322_groups_0, pad = var_5322_pad_0, pad_type = var_5322_pad_type_0, strides = var_5322_strides_0, weight = layers_13_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_361_cast_fp16)[name = tensor("op_5322_cast_fp16")]; + tensor var_5328_pad_type_0 = const()[name = tensor("op_5328_pad_type_0"), val = tensor("valid")]; + tensor var_5328_strides_0 = const()[name = tensor("op_5328_strides_0"), val = tensor([1, 1])]; + tensor var_5328_pad_0 = const()[name = tensor("op_5328_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5328_dilations_0 = const()[name = tensor("op_5328_dilations_0"), val = tensor([1, 1])]; + tensor var_5328_groups_0 = const()[name = tensor("op_5328_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316531008))), name = tensor("layers_13_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316499264))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5328_cast_fp16 = conv(dilations = var_5328_dilations_0, groups = var_5328_groups_0, pad = var_5328_pad_0, pad_type = var_5328_pad_type_0, strides = var_5328_strides_0, weight = layers_13_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_361_cast_fp16)[name = tensor("op_5328_cast_fp16")]; + tensor obj_57_cast_fp16 = add(x = var_5322_cast_fp16, y = var_5328_cast_fp16)[name = tensor("obj_57_cast_fp16")]; + tensor inputs_135_cast_fp16 = add(x = inputs_133_cast_fp16, y = obj_57_cast_fp16)[name = tensor("inputs_135_cast_fp16")]; + tensor out_135_axes_0 = const()[name = tensor("out_135_axes_0"), val = tensor([1])]; + tensor var_5339_to_fp16 = const()[name = tensor("op_5339_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_135_cast_fp16 = layer_norm(axes = out_135_axes_0, epsilon = var_5339_to_fp16, x = inputs_135_cast_fp16)[name = tensor("out_135_cast_fp16")]; + tensor input_363_gamma_0_to_fp16 = const()[name = tensor("input_363_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316662144)))]; + tensor input_363_beta_0_to_fp16 = const()[name = tensor("input_363_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316664256)))]; + tensor input_363_epsilon_0_to_fp16 = const()[name = tensor("input_363_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_363_cast_fp16 = batch_norm(beta = input_363_beta_0_to_fp16, epsilon = input_363_epsilon_0_to_fp16, gamma = input_363_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_135_cast_fp16)[name = tensor("input_363_cast_fp16")]; + tensor var_5360_pad_type_0 = const()[name = tensor("op_5360_pad_type_0"), val = tensor("valid")]; + tensor var_5360_strides_0 = const()[name = tensor("op_5360_strides_0"), val = tensor([1, 1])]; + tensor var_5360_pad_0 = const()[name = tensor("op_5360_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5360_dilations_0 = const()[name = tensor("op_5360_dilations_0"), val = tensor([1, 1])]; + tensor var_5360_groups_0 = const()[name = tensor("op_5360_groups_0"), val = tensor(1)]; + tensor layers_13_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316666368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(318239296))), name = tensor("layers_13_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_5360_cast_fp16 = conv(dilations = var_5360_dilations_0, groups = var_5360_groups_0, pad = var_5360_pad_0, pad_type = var_5360_pad_type_0, strides = var_5360_strides_0, weight = layers_13_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_363_cast_fp16)[name = tensor("op_5360_cast_fp16")]; + tensor var_5366_pad_type_0 = const()[name = tensor("op_5366_pad_type_0"), val = tensor("valid")]; + tensor var_5366_strides_0 = const()[name = tensor("op_5366_strides_0"), val = tensor([1, 1])]; + tensor var_5366_pad_0 = const()[name = tensor("op_5366_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5366_dilations_0 = const()[name = tensor("op_5366_dilations_0"), val = tensor([1, 1])]; + tensor var_5366_groups_0 = const()[name = tensor("op_5366_groups_0"), val = tensor(1)]; + tensor layers_13_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(318301888))), name = tensor("layers_13_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(318239488))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_5366_cast_fp16 = conv(dilations = var_5366_dilations_0, groups = var_5366_groups_0, pad = var_5366_pad_0, pad_type = var_5366_pad_type_0, strides = var_5366_strides_0, weight = layers_13_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_363_cast_fp16)[name = tensor("op_5366_cast_fp16")]; + tensor input_365_cast_fp16 = add(x = var_5360_cast_fp16, y = var_5366_cast_fp16)[name = tensor("input_365_cast_fp16")]; + tensor input_367_split_num_splits_0 = const()[name = tensor("input_367_split_num_splits_0"), val = tensor(2)]; + tensor input_367_split_axis_0 = const()[name = tensor("input_367_split_axis_0"), val = tensor(1)]; + tensor input_367_split_cast_fp16_0, tensor input_367_split_cast_fp16_1 = split(axis = input_367_split_axis_0, num_splits = input_367_split_num_splits_0, x = input_365_cast_fp16)[name = tensor("input_367_split_cast_fp16")]; + tensor input_367_split_1_sigmoid_cast_fp16 = sigmoid(x = input_367_split_cast_fp16_1)[name = tensor("input_367_split_1_sigmoid_cast_fp16")]; + tensor input_367_cast_fp16 = mul(x = input_367_split_cast_fp16_0, y = input_367_split_1_sigmoid_cast_fp16)[name = tensor("input_367_cast_fp16")]; + tensor input_369_pad_type_0 = const()[name = tensor("input_369_pad_type_0"), val = tensor("custom")]; + tensor input_369_pad_0 = const()[name = tensor("input_369_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_369_groups_0 = const()[name = tensor("input_369_groups_0"), val = tensor(1024)]; + tensor input_369_strides_0 = const()[name = tensor("input_369_strides_0"), val = tensor([1, 1])]; + tensor input_369_dilations_0 = const()[name = tensor("input_369_dilations_0"), val = tensor([1, 1])]; + tensor const_294_to_fp16 = const()[name = tensor("const_294_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(318564096)))]; + tensor const_295_to_fp16 = const()[name = tensor("const_295_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(318582592)))]; + tensor input_371_cast_fp16 = conv(bias = const_295_to_fp16, dilations = input_369_dilations_0, groups = input_369_groups_0, pad = input_369_pad_0, pad_type = input_369_pad_type_0, strides = input_369_strides_0, weight = const_294_to_fp16, x = input_367_cast_fp16)[name = tensor("input_371_cast_fp16")]; + tensor input_373_cast_fp16 = silu(x = input_371_cast_fp16)[name = tensor("input_373_cast_fp16")]; + tensor var_5388_pad_type_0 = const()[name = tensor("op_5388_pad_type_0"), val = tensor("valid")]; + tensor var_5388_strides_0 = const()[name = tensor("op_5388_strides_0"), val = tensor([1, 1])]; + tensor var_5388_pad_0 = const()[name = tensor("op_5388_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5388_dilations_0 = const()[name = tensor("op_5388_dilations_0"), val = tensor([1, 1])]; + tensor var_5388_groups_0 = const()[name = tensor("op_5388_groups_0"), val = tensor(1)]; + tensor layers_13_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(318584704))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(319371200))), name = tensor("layers_13_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5388_cast_fp16 = conv(dilations = var_5388_dilations_0, groups = var_5388_groups_0, pad = var_5388_pad_0, pad_type = var_5388_pad_type_0, strides = var_5388_strides_0, weight = layers_13_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_373_cast_fp16)[name = tensor("op_5388_cast_fp16")]; + tensor var_5394_pad_type_0 = const()[name = tensor("op_5394_pad_type_0"), val = tensor("valid")]; + tensor var_5394_strides_0 = const()[name = tensor("op_5394_strides_0"), val = tensor([1, 1])]; + tensor var_5394_pad_0 = const()[name = tensor("op_5394_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5394_dilations_0 = const()[name = tensor("op_5394_dilations_0"), val = tensor([1, 1])]; + tensor var_5394_groups_0 = const()[name = tensor("op_5394_groups_0"), val = tensor(1)]; + tensor layers_13_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(319403072))), name = tensor("layers_13_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(319371392))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5394_cast_fp16 = conv(dilations = var_5394_dilations_0, groups = var_5394_groups_0, pad = var_5394_pad_0, pad_type = var_5394_pad_type_0, strides = var_5394_strides_0, weight = layers_13_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_373_cast_fp16)[name = tensor("op_5394_cast_fp16")]; + tensor x_83_cast_fp16 = add(x = var_5388_cast_fp16, y = var_5394_cast_fp16)[name = tensor("x_83_cast_fp16")]; + tensor inputs_137_cast_fp16 = add(x = inputs_135_cast_fp16, y = x_83_cast_fp16)[name = tensor("inputs_137_cast_fp16")]; + tensor out_137_axes_0 = const()[name = tensor("out_137_axes_0"), val = tensor([1])]; + tensor var_5405_to_fp16 = const()[name = tensor("op_5405_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_137_cast_fp16 = layer_norm(axes = out_137_axes_0, epsilon = var_5405_to_fp16, x = inputs_137_cast_fp16)[name = tensor("out_137_cast_fp16")]; + tensor input_375_gamma_0_to_fp16 = const()[name = tensor("input_375_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(319534208)))]; + tensor input_375_beta_0_to_fp16 = const()[name = tensor("input_375_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(319536320)))]; + tensor input_375_epsilon_0_to_fp16 = const()[name = tensor("input_375_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_375_cast_fp16 = batch_norm(beta = input_375_beta_0_to_fp16, epsilon = input_375_epsilon_0_to_fp16, gamma = input_375_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_137_cast_fp16)[name = tensor("input_375_cast_fp16")]; + tensor var_5425_pad_type_0 = const()[name = tensor("op_5425_pad_type_0"), val = tensor("valid")]; + tensor var_5425_strides_0 = const()[name = tensor("op_5425_strides_0"), val = tensor([1, 1])]; + tensor var_5425_pad_0 = const()[name = tensor("op_5425_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5425_dilations_0 = const()[name = tensor("op_5425_dilations_0"), val = tensor([1, 1])]; + tensor var_5425_groups_0 = const()[name = tensor("op_5425_groups_0"), val = tensor(1)]; + tensor layers_13_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(319538432))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(322684224))), name = tensor("layers_13_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_5425_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_5425_dilations_0, groups = var_5425_groups_0, pad = var_5425_pad_0, pad_type = var_5425_pad_type_0, strides = var_5425_strides_0, weight = layers_13_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_375_cast_fp16)[name = tensor("op_5425_cast_fp16")]; + tensor var_5431_pad_type_0 = const()[name = tensor("op_5431_pad_type_0"), val = tensor("valid")]; + tensor var_5431_strides_0 = const()[name = tensor("op_5431_strides_0"), val = tensor([1, 1])]; + tensor var_5431_pad_0 = const()[name = tensor("op_5431_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5431_dilations_0 = const()[name = tensor("op_5431_dilations_0"), val = tensor([1, 1])]; + tensor var_5431_groups_0 = const()[name = tensor("op_5431_groups_0"), val = tensor(1)]; + tensor layers_13_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(322839232))), name = tensor("layers_13_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(322684416))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_5431_cast_fp16 = conv(dilations = var_5431_dilations_0, groups = var_5431_groups_0, pad = var_5431_pad_0, pad_type = var_5431_pad_type_0, strides = var_5431_strides_0, weight = layers_13_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_375_cast_fp16)[name = tensor("op_5431_cast_fp16")]; + tensor input_377_cast_fp16 = add(x = var_5425_cast_fp16, y = var_5431_cast_fp16)[name = tensor("input_377_cast_fp16")]; + tensor input_379_cast_fp16 = silu(x = input_377_cast_fp16)[name = tensor("input_379_cast_fp16")]; + tensor var_5442_pad_type_0 = const()[name = tensor("op_5442_pad_type_0"), val = tensor("valid")]; + tensor var_5442_strides_0 = const()[name = tensor("op_5442_strides_0"), val = tensor([1, 1])]; + tensor var_5442_pad_0 = const()[name = tensor("op_5442_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5442_dilations_0 = const()[name = tensor("op_5442_dilations_0"), val = tensor([1, 1])]; + tensor var_5442_groups_0 = const()[name = tensor("op_5442_groups_0"), val = tensor(1)]; + tensor layers_13_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(323363584))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(326509376))), name = tensor("layers_13_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_5442_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5442_dilations_0, groups = var_5442_groups_0, pad = var_5442_pad_0, pad_type = var_5442_pad_type_0, strides = var_5442_strides_0, weight = layers_13_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_379_cast_fp16)[name = tensor("op_5442_cast_fp16")]; + tensor var_5448_pad_type_0 = const()[name = tensor("op_5448_pad_type_0"), val = tensor("valid")]; + tensor var_5448_strides_0 = const()[name = tensor("op_5448_strides_0"), val = tensor([1, 1])]; + tensor var_5448_pad_0 = const()[name = tensor("op_5448_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5448_dilations_0 = const()[name = tensor("op_5448_dilations_0"), val = tensor([1, 1])]; + tensor var_5448_groups_0 = const()[name = tensor("op_5448_groups_0"), val = tensor(1)]; + tensor layers_13_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(326674304))), name = tensor("layers_13_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(326509568))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_5448_cast_fp16 = conv(dilations = var_5448_dilations_0, groups = var_5448_groups_0, pad = var_5448_pad_0, pad_type = var_5448_pad_type_0, strides = var_5448_strides_0, weight = layers_13_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_379_cast_fp16)[name = tensor("op_5448_cast_fp16")]; + tensor x_85_cast_fp16 = add(x = var_5442_cast_fp16, y = var_5448_cast_fp16)[name = tensor("x_85_cast_fp16")]; + tensor var_5450_to_fp16 = const()[name = tensor("op_5450_to_fp16"), val = tensor(0x1p-1)]; + tensor var_5451_cast_fp16 = mul(x = x_85_cast_fp16, y = var_5450_to_fp16)[name = tensor("op_5451_cast_fp16")]; + tensor inputs_139_cast_fp16 = add(x = inputs_137_cast_fp16, y = var_5451_cast_fp16)[name = tensor("inputs_139_cast_fp16")]; + tensor out_139_axes_0 = const()[name = tensor("out_139_axes_0"), val = tensor([1])]; + tensor var_5461_to_fp16 = const()[name = tensor("op_5461_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_139_cast_fp16 = layer_norm(axes = out_139_axes_0, epsilon = var_5461_to_fp16, x = inputs_139_cast_fp16)[name = tensor("out_139_cast_fp16")]; + tensor inputs_141_gamma_0_to_fp16 = const()[name = tensor("inputs_141_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(327198656)))]; + tensor inputs_141_beta_0_to_fp16 = const()[name = tensor("inputs_141_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(327200768)))]; + tensor inputs_141_epsilon_0_to_fp16 = const()[name = tensor("inputs_141_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_141_cast_fp16 = batch_norm(beta = inputs_141_beta_0_to_fp16, epsilon = inputs_141_epsilon_0_to_fp16, gamma = inputs_141_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_139_cast_fp16)[name = tensor("inputs_141_cast_fp16")]; + tensor var_5475 = const()[name = tensor("op_5475"), val = tensor(3)]; + tensor out_141_axes_0 = const()[name = tensor("out_141_axes_0"), val = tensor([1])]; + tensor var_5506_to_fp16 = const()[name = tensor("op_5506_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_141_cast_fp16 = layer_norm(axes = out_141_axes_0, epsilon = var_5506_to_fp16, x = inputs_141_cast_fp16)[name = tensor("out_141_cast_fp16")]; + tensor input_381_gamma_0_to_fp16 = const()[name = tensor("input_381_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(327202880)))]; + tensor input_381_beta_0_to_fp16 = const()[name = tensor("input_381_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(327204992)))]; + tensor input_381_epsilon_0_to_fp16 = const()[name = tensor("input_381_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_381_cast_fp16 = batch_norm(beta = input_381_beta_0_to_fp16, epsilon = input_381_epsilon_0_to_fp16, gamma = input_381_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_141_cast_fp16)[name = tensor("input_381_cast_fp16")]; + tensor var_5526_pad_type_0 = const()[name = tensor("op_5526_pad_type_0"), val = tensor("valid")]; + tensor var_5526_strides_0 = const()[name = tensor("op_5526_strides_0"), val = tensor([1, 1])]; + tensor var_5526_pad_0 = const()[name = tensor("op_5526_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5526_dilations_0 = const()[name = tensor("op_5526_dilations_0"), val = tensor([1, 1])]; + tensor var_5526_groups_0 = const()[name = tensor("op_5526_groups_0"), val = tensor(1)]; + tensor layers_14_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(327207104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(330352896))), name = tensor("layers_14_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_5526_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_5526_dilations_0, groups = var_5526_groups_0, pad = var_5526_pad_0, pad_type = var_5526_pad_type_0, strides = var_5526_strides_0, weight = layers_14_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_381_cast_fp16)[name = tensor("op_5526_cast_fp16")]; + tensor var_5532_pad_type_0 = const()[name = tensor("op_5532_pad_type_0"), val = tensor("valid")]; + tensor var_5532_strides_0 = const()[name = tensor("op_5532_strides_0"), val = tensor([1, 1])]; + tensor var_5532_pad_0 = const()[name = tensor("op_5532_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5532_dilations_0 = const()[name = tensor("op_5532_dilations_0"), val = tensor([1, 1])]; + tensor var_5532_groups_0 = const()[name = tensor("op_5532_groups_0"), val = tensor(1)]; + tensor layers_14_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(330532352))), name = tensor("layers_14_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(330353088))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_5532_cast_fp16 = conv(dilations = var_5532_dilations_0, groups = var_5532_groups_0, pad = var_5532_pad_0, pad_type = var_5532_pad_type_0, strides = var_5532_strides_0, weight = layers_14_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_381_cast_fp16)[name = tensor("op_5532_cast_fp16")]; + tensor input_383_cast_fp16 = add(x = var_5526_cast_fp16, y = var_5532_cast_fp16)[name = tensor("input_383_cast_fp16")]; + tensor input_385_cast_fp16 = silu(x = input_383_cast_fp16)[name = tensor("input_385_cast_fp16")]; + tensor var_5543_pad_type_0 = const()[name = tensor("op_5543_pad_type_0"), val = tensor("valid")]; + tensor var_5543_strides_0 = const()[name = tensor("op_5543_strides_0"), val = tensor([1, 1])]; + tensor var_5543_pad_0 = const()[name = tensor("op_5543_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5543_dilations_0 = const()[name = tensor("op_5543_dilations_0"), val = tensor([1, 1])]; + tensor var_5543_groups_0 = const()[name = tensor("op_5543_groups_0"), val = tensor(1)]; + tensor layers_14_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(331056704))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(334202496))), name = tensor("layers_14_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_5543_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5543_dilations_0, groups = var_5543_groups_0, pad = var_5543_pad_0, pad_type = var_5543_pad_type_0, strides = var_5543_strides_0, weight = layers_14_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_385_cast_fp16)[name = tensor("op_5543_cast_fp16")]; + tensor var_5549_pad_type_0 = const()[name = tensor("op_5549_pad_type_0"), val = tensor("valid")]; + tensor var_5549_strides_0 = const()[name = tensor("op_5549_strides_0"), val = tensor([1, 1])]; + tensor var_5549_pad_0 = const()[name = tensor("op_5549_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5549_dilations_0 = const()[name = tensor("op_5549_dilations_0"), val = tensor([1, 1])]; + tensor var_5549_groups_0 = const()[name = tensor("op_5549_groups_0"), val = tensor(1)]; + tensor layers_14_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(334378432))), name = tensor("layers_14_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(334202688))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_5549_cast_fp16 = conv(dilations = var_5549_dilations_0, groups = var_5549_groups_0, pad = var_5549_pad_0, pad_type = var_5549_pad_type_0, strides = var_5549_strides_0, weight = layers_14_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_385_cast_fp16)[name = tensor("op_5549_cast_fp16")]; + tensor x_87_cast_fp16 = add(x = var_5543_cast_fp16, y = var_5549_cast_fp16)[name = tensor("x_87_cast_fp16")]; + tensor var_5551_to_fp16 = const()[name = tensor("op_5551_to_fp16"), val = tensor(0x1p-1)]; + tensor var_5552_cast_fp16 = mul(x = x_87_cast_fp16, y = var_5551_to_fp16)[name = tensor("op_5552_cast_fp16")]; + tensor inputs_143_cast_fp16 = add(x = inputs_141_cast_fp16, y = var_5552_cast_fp16)[name = tensor("inputs_143_cast_fp16")]; + tensor out_143_axes_0 = const()[name = tensor("out_143_axes_0"), val = tensor([1])]; + tensor var_5562_to_fp16 = const()[name = tensor("op_5562_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_143_cast_fp16 = layer_norm(axes = out_143_axes_0, epsilon = var_5562_to_fp16, x = inputs_143_cast_fp16)[name = tensor("out_143_cast_fp16")]; + tensor obj_59_gamma_0_to_fp16 = const()[name = tensor("obj_59_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(334902784)))]; + tensor obj_59_beta_0_to_fp16 = const()[name = tensor("obj_59_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(334904896)))]; + tensor obj_59_epsilon_0_to_fp16 = const()[name = tensor("obj_59_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_59_cast_fp16 = batch_norm(beta = obj_59_beta_0_to_fp16, epsilon = obj_59_epsilon_0_to_fp16, gamma = obj_59_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_143_cast_fp16)[name = tensor("obj_59_cast_fp16")]; + tensor var_5587_pad_type_0 = const()[name = tensor("op_5587_pad_type_0"), val = tensor("valid")]; + tensor var_5587_strides_0 = const()[name = tensor("op_5587_strides_0"), val = tensor([1, 1])]; + tensor var_5587_pad_0 = const()[name = tensor("op_5587_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5587_dilations_0 = const()[name = tensor("op_5587_dilations_0"), val = tensor([1, 1])]; + tensor var_5587_groups_0 = const()[name = tensor("op_5587_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(334907008))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335693504))), name = tensor("layers_14_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5587_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5587_dilations_0, groups = var_5587_groups_0, pad = var_5587_pad_0, pad_type = var_5587_pad_type_0, strides = var_5587_strides_0, weight = layers_14_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_59_cast_fp16)[name = tensor("op_5587_cast_fp16")]; + tensor var_5593_pad_type_0 = const()[name = tensor("op_5593_pad_type_0"), val = tensor("valid")]; + tensor var_5593_strides_0 = const()[name = tensor("op_5593_strides_0"), val = tensor([1, 1])]; + tensor var_5593_pad_0 = const()[name = tensor("op_5593_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5593_dilations_0 = const()[name = tensor("op_5593_dilations_0"), val = tensor([1, 1])]; + tensor var_5593_groups_0 = const()[name = tensor("op_5593_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335727296))), name = tensor("layers_14_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335693696))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5593_cast_fp16 = conv(dilations = var_5593_dilations_0, groups = var_5593_groups_0, pad = var_5593_pad_0, pad_type = var_5593_pad_type_0, strides = var_5593_strides_0, weight = layers_14_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_59_cast_fp16)[name = tensor("op_5593_cast_fp16")]; + tensor query_57_cast_fp16 = add(x = var_5587_cast_fp16, y = var_5593_cast_fp16)[name = tensor("query_57_cast_fp16")]; + tensor var_5602_pad_type_0 = const()[name = tensor("op_5602_pad_type_0"), val = tensor("valid")]; + tensor var_5602_strides_0 = const()[name = tensor("op_5602_strides_0"), val = tensor([1, 1])]; + tensor var_5602_pad_0 = const()[name = tensor("op_5602_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5602_dilations_0 = const()[name = tensor("op_5602_dilations_0"), val = tensor([1, 1])]; + tensor var_5602_groups_0 = const()[name = tensor("op_5602_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335858432))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(336644928))), name = tensor("layers_14_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5602_cast_fp16 = conv(dilations = var_5602_dilations_0, groups = var_5602_groups_0, pad = var_5602_pad_0, pad_type = var_5602_pad_type_0, strides = var_5602_strides_0, weight = layers_14_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_59_cast_fp16)[name = tensor("op_5602_cast_fp16")]; + tensor var_5608_pad_type_0 = const()[name = tensor("op_5608_pad_type_0"), val = tensor("valid")]; + tensor var_5608_strides_0 = const()[name = tensor("op_5608_strides_0"), val = tensor([1, 1])]; + tensor var_5608_pad_0 = const()[name = tensor("op_5608_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5608_dilations_0 = const()[name = tensor("op_5608_dilations_0"), val = tensor([1, 1])]; + tensor var_5608_groups_0 = const()[name = tensor("op_5608_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(336683968))), name = tensor("layers_14_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(336645120))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5608_cast_fp16 = conv(dilations = var_5608_dilations_0, groups = var_5608_groups_0, pad = var_5608_pad_0, pad_type = var_5608_pad_type_0, strides = var_5608_strides_0, weight = layers_14_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_59_cast_fp16)[name = tensor("op_5608_cast_fp16")]; + tensor key_29_cast_fp16 = add(x = var_5602_cast_fp16, y = var_5608_cast_fp16)[name = tensor("key_29_cast_fp16")]; + tensor var_5618_pad_type_0 = const()[name = tensor("op_5618_pad_type_0"), val = tensor("valid")]; + tensor var_5618_strides_0 = const()[name = tensor("op_5618_strides_0"), val = tensor([1, 1])]; + tensor var_5618_pad_0 = const()[name = tensor("op_5618_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5618_dilations_0 = const()[name = tensor("op_5618_dilations_0"), val = tensor([1, 1])]; + tensor var_5618_groups_0 = const()[name = tensor("op_5618_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(336815104))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337601600))), name = tensor("layers_14_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5618_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5618_dilations_0, groups = var_5618_groups_0, pad = var_5618_pad_0, pad_type = var_5618_pad_type_0, strides = var_5618_strides_0, weight = layers_14_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_59_cast_fp16)[name = tensor("op_5618_cast_fp16")]; + tensor var_5624_pad_type_0 = const()[name = tensor("op_5624_pad_type_0"), val = tensor("valid")]; + tensor var_5624_strides_0 = const()[name = tensor("op_5624_strides_0"), val = tensor([1, 1])]; + tensor var_5624_pad_0 = const()[name = tensor("op_5624_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5624_dilations_0 = const()[name = tensor("op_5624_dilations_0"), val = tensor([1, 1])]; + tensor var_5624_groups_0 = const()[name = tensor("op_5624_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337634560))), name = tensor("layers_14_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337601792))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5624_cast_fp16 = conv(dilations = var_5624_dilations_0, groups = var_5624_groups_0, pad = var_5624_pad_0, pad_type = var_5624_pad_type_0, strides = var_5624_strides_0, weight = layers_14_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_59_cast_fp16)[name = tensor("op_5624_cast_fp16")]; + tensor value_29_cast_fp16 = add(x = var_5618_cast_fp16, y = var_5624_cast_fp16)[name = tensor("value_29_cast_fp16")]; + tensor var_5627_to_fp16 = const()[name = tensor("op_5627_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337765696)))]; + tensor query_59_cast_fp16 = add(x = query_57_cast_fp16, y = var_5627_to_fp16)[name = tensor("query_59_cast_fp16")]; + tensor var_5630_to_fp16 = const()[name = tensor("op_5630_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337767808)))]; + tensor q_with_bias_v_29_cast_fp16 = add(x = query_57_cast_fp16, y = var_5630_to_fp16)[name = tensor("q_with_bias_v_29_cast_fp16")]; + tensor var_5640_pad_type_0 = const()[name = tensor("op_5640_pad_type_0"), val = tensor("valid")]; + tensor var_5640_strides_0 = const()[name = tensor("op_5640_strides_0"), val = tensor([1, 1])]; + tensor var_5640_pad_0 = const()[name = tensor("op_5640_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5640_dilations_0 = const()[name = tensor("op_5640_dilations_0"), val = tensor([1, 1])]; + tensor var_5640_groups_0 = const()[name = tensor("op_5640_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337769920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(338556416))), name = tensor("layers_14_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5640_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5640_dilations_0, groups = var_5640_groups_0, pad = var_5640_pad_0, pad_type = var_5640_pad_type_0, strides = var_5640_strides_0, weight = layers_14_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_5640_cast_fp16")]; + tensor var_5646_pad_type_0 = const()[name = tensor("op_5646_pad_type_0"), val = tensor("valid")]; + tensor var_5646_strides_0 = const()[name = tensor("op_5646_strides_0"), val = tensor([1, 1])]; + tensor var_5646_pad_0 = const()[name = tensor("op_5646_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5646_dilations_0 = const()[name = tensor("op_5646_dilations_0"), val = tensor([1, 1])]; + tensor var_5646_groups_0 = const()[name = tensor("op_5646_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(338621824))), name = tensor("layers_14_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(338556608))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5646_cast_fp16 = conv(dilations = var_5646_dilations_0, groups = var_5646_groups_0, pad = var_5646_pad_0, pad_type = var_5646_pad_type_0, strides = var_5646_strides_0, weight = layers_14_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_5646_cast_fp16")]; + tensor p_29_cast_fp16 = add(x = var_5640_cast_fp16, y = var_5646_cast_fp16)[name = tensor("p_29_cast_fp16")]; + tensor var_5650 = const()[name = tensor("op_5650"), val = tensor([1, 8, 128, 188])]; + tensor var_5651_cast_fp16 = reshape(shape = var_5650, x = q_with_bias_v_29_cast_fp16)[name = tensor("op_5651_cast_fp16")]; + tensor var_5652 = const()[name = tensor("op_5652"), val = tensor([1, 8, 128, -1])]; + tensor var_5653_cast_fp16 = reshape(shape = var_5652, x = p_29_cast_fp16)[name = tensor("op_5653_cast_fp16")]; + tensor matrix_bd_113_transpose_x_0 = const()[name = tensor("matrix_bd_113_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_113_transpose_y_0 = const()[name = tensor("matrix_bd_113_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_113_cast_fp16 = matmul(transpose_x = matrix_bd_113_transpose_x_0, transpose_y = matrix_bd_113_transpose_y_0, x = var_5651_cast_fp16, y = var_5653_cast_fp16)[name = tensor("matrix_bd_113_cast_fp16")]; + tensor matrix_bd_115_pad_0 = const()[name = tensor("matrix_bd_115_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_115_mode_0 = const()[name = tensor("matrix_bd_115_mode_0"), val = tensor("constant")]; + tensor const_164_to_fp16 = const()[name = tensor("const_164_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_115_cast_fp16 = pad(constant_val = const_164_to_fp16, mode = matrix_bd_115_mode_0, pad = matrix_bd_115_pad_0, x = matrix_bd_113_cast_fp16)[name = tensor("matrix_bd_115_cast_fp16")]; + tensor var_5662 = const()[name = tensor("op_5662"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_117_cast_fp16 = reshape(shape = var_5662, x = matrix_bd_115_cast_fp16)[name = tensor("matrix_bd_117_cast_fp16")]; + tensor var_5666_begin_0 = const()[name = tensor("op_5666_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_5666_end_0 = const()[name = tensor("op_5666_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_5666_end_mask_0 = const()[name = tensor("op_5666_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_5666_cast_fp16 = slice_by_index(begin = var_5666_begin_0, end = var_5666_end_0, end_mask = var_5666_end_mask_0, x = matrix_bd_117_cast_fp16)[name = tensor("op_5666_cast_fp16")]; + tensor var_5667 = const()[name = tensor("op_5667"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_119_cast_fp16 = reshape(shape = var_5667, x = var_5666_cast_fp16)[name = tensor("matrix_bd_119_cast_fp16")]; + tensor var_5672_begin_0 = const()[name = tensor("op_5672_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5672_end_0 = const()[name = tensor("op_5672_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_5672_end_mask_0 = const()[name = tensor("op_5672_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5672_cast_fp16 = slice_by_index(begin = var_5672_begin_0, end = var_5672_end_0, end_mask = var_5672_end_mask_0, x = matrix_bd_119_cast_fp16)[name = tensor("op_5672_cast_fp16")]; + tensor var_5673_to_fp16 = const()[name = tensor("op_5673_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_29_cast_fp16 = mul(x = var_5672_cast_fp16, y = var_5673_to_fp16)[name = tensor("qk_mask_29_cast_fp16")]; + tensor var_5677 = const()[name = tensor("op_5677"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_29_cast_fp16 = reshape(shape = var_5677, x = query_59_cast_fp16)[name = tensor("mh_q_29_cast_fp16")]; + tensor var_5679_to_fp16 = const()[name = tensor("op_5679_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_5680_cast_fp16 = mul(x = mh_q_29_cast_fp16, y = var_5679_to_fp16)[name = tensor("op_5680_cast_fp16")]; + tensor var_5683 = const()[name = tensor("op_5683"), val = tensor([1, 8, 128, 188])]; + tensor var_5684_cast_fp16 = reshape(shape = var_5683, x = key_29_cast_fp16)[name = tensor("op_5684_cast_fp16")]; + tensor mh_w_57_transpose_x_0 = const()[name = tensor("mh_w_57_transpose_x_0"), val = tensor(true)]; + tensor mh_w_57_transpose_y_0 = const()[name = tensor("mh_w_57_transpose_y_0"), val = tensor(false)]; + tensor mh_w_57_cast_fp16 = matmul(transpose_x = mh_w_57_transpose_x_0, transpose_y = mh_w_57_transpose_y_0, x = var_5680_cast_fp16, y = var_5684_cast_fp16)[name = tensor("mh_w_57_cast_fp16")]; + tensor mh_w_59_cast_fp16 = add(x = mh_w_57_cast_fp16, y = qk_mask_29_cast_fp16)[name = tensor("mh_w_59_cast_fp16")]; + tensor var_5688_cast_fp16 = softmax(axis = var_5475, x = mh_w_59_cast_fp16)[name = tensor("op_5688_cast_fp16")]; + tensor var_5689 = const()[name = tensor("op_5689"), val = tensor([1, 8, 128, 188])]; + tensor var_5690_cast_fp16 = reshape(shape = var_5689, x = value_29_cast_fp16)[name = tensor("op_5690_cast_fp16")]; + tensor attn_29_transpose_x_0 = const()[name = tensor("attn_29_transpose_x_0"), val = tensor(false)]; + tensor attn_29_transpose_y_0 = const()[name = tensor("attn_29_transpose_y_0"), val = tensor(true)]; + tensor attn_29_cast_fp16 = matmul(transpose_x = attn_29_transpose_x_0, transpose_y = attn_29_transpose_y_0, x = var_5690_cast_fp16, y = var_5688_cast_fp16)[name = tensor("attn_29_cast_fp16")]; + tensor var_5693 = const()[name = tensor("op_5693"), val = tensor([1, 1024, 1, 188])]; + tensor input_387_cast_fp16 = reshape(shape = var_5693, x = attn_29_cast_fp16)[name = tensor("input_387_cast_fp16")]; + tensor var_5703_pad_type_0 = const()[name = tensor("op_5703_pad_type_0"), val = tensor("valid")]; + tensor var_5703_strides_0 = const()[name = tensor("op_5703_strides_0"), val = tensor([1, 1])]; + tensor var_5703_pad_0 = const()[name = tensor("op_5703_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5703_dilations_0 = const()[name = tensor("op_5703_dilations_0"), val = tensor([1, 1])]; + tensor var_5703_groups_0 = const()[name = tensor("op_5703_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(338752960))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339539456))), name = tensor("layers_14_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5703_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5703_dilations_0, groups = var_5703_groups_0, pad = var_5703_pad_0, pad_type = var_5703_pad_type_0, strides = var_5703_strides_0, weight = layers_14_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_387_cast_fp16)[name = tensor("op_5703_cast_fp16")]; + tensor var_5709_pad_type_0 = const()[name = tensor("op_5709_pad_type_0"), val = tensor("valid")]; + tensor var_5709_strides_0 = const()[name = tensor("op_5709_strides_0"), val = tensor([1, 1])]; + tensor var_5709_pad_0 = const()[name = tensor("op_5709_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5709_dilations_0 = const()[name = tensor("op_5709_dilations_0"), val = tensor([1, 1])]; + tensor var_5709_groups_0 = const()[name = tensor("op_5709_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339573120))), name = tensor("layers_14_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339539648))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5709_cast_fp16 = conv(dilations = var_5709_dilations_0, groups = var_5709_groups_0, pad = var_5709_pad_0, pad_type = var_5709_pad_type_0, strides = var_5709_strides_0, weight = layers_14_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_387_cast_fp16)[name = tensor("op_5709_cast_fp16")]; + tensor obj_61_cast_fp16 = add(x = var_5703_cast_fp16, y = var_5709_cast_fp16)[name = tensor("obj_61_cast_fp16")]; + tensor inputs_145_cast_fp16 = add(x = inputs_143_cast_fp16, y = obj_61_cast_fp16)[name = tensor("inputs_145_cast_fp16")]; + tensor out_145_axes_0 = const()[name = tensor("out_145_axes_0"), val = tensor([1])]; + tensor var_5720_to_fp16 = const()[name = tensor("op_5720_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_145_cast_fp16 = layer_norm(axes = out_145_axes_0, epsilon = var_5720_to_fp16, x = inputs_145_cast_fp16)[name = tensor("out_145_cast_fp16")]; + tensor input_389_gamma_0_to_fp16 = const()[name = tensor("input_389_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339704256)))]; + tensor input_389_beta_0_to_fp16 = const()[name = tensor("input_389_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339706368)))]; + tensor input_389_epsilon_0_to_fp16 = const()[name = tensor("input_389_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_389_cast_fp16 = batch_norm(beta = input_389_beta_0_to_fp16, epsilon = input_389_epsilon_0_to_fp16, gamma = input_389_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_145_cast_fp16)[name = tensor("input_389_cast_fp16")]; + tensor var_5741_pad_type_0 = const()[name = tensor("op_5741_pad_type_0"), val = tensor("valid")]; + tensor var_5741_strides_0 = const()[name = tensor("op_5741_strides_0"), val = tensor([1, 1])]; + tensor var_5741_pad_0 = const()[name = tensor("op_5741_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5741_dilations_0 = const()[name = tensor("op_5741_dilations_0"), val = tensor([1, 1])]; + tensor var_5741_groups_0 = const()[name = tensor("op_5741_groups_0"), val = tensor(1)]; + tensor layers_14_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339708480))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(341281408))), name = tensor("layers_14_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_5741_cast_fp16 = conv(dilations = var_5741_dilations_0, groups = var_5741_groups_0, pad = var_5741_pad_0, pad_type = var_5741_pad_type_0, strides = var_5741_strides_0, weight = layers_14_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_389_cast_fp16)[name = tensor("op_5741_cast_fp16")]; + tensor var_5747_pad_type_0 = const()[name = tensor("op_5747_pad_type_0"), val = tensor("valid")]; + tensor var_5747_strides_0 = const()[name = tensor("op_5747_strides_0"), val = tensor([1, 1])]; + tensor var_5747_pad_0 = const()[name = tensor("op_5747_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5747_dilations_0 = const()[name = tensor("op_5747_dilations_0"), val = tensor([1, 1])]; + tensor var_5747_groups_0 = const()[name = tensor("op_5747_groups_0"), val = tensor(1)]; + tensor layers_14_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(341344512))), name = tensor("layers_14_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(341281600))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_5747_cast_fp16 = conv(dilations = var_5747_dilations_0, groups = var_5747_groups_0, pad = var_5747_pad_0, pad_type = var_5747_pad_type_0, strides = var_5747_strides_0, weight = layers_14_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_389_cast_fp16)[name = tensor("op_5747_cast_fp16")]; + tensor input_391_cast_fp16 = add(x = var_5741_cast_fp16, y = var_5747_cast_fp16)[name = tensor("input_391_cast_fp16")]; + tensor input_393_split_num_splits_0 = const()[name = tensor("input_393_split_num_splits_0"), val = tensor(2)]; + tensor input_393_split_axis_0 = const()[name = tensor("input_393_split_axis_0"), val = tensor(1)]; + tensor input_393_split_cast_fp16_0, tensor input_393_split_cast_fp16_1 = split(axis = input_393_split_axis_0, num_splits = input_393_split_num_splits_0, x = input_391_cast_fp16)[name = tensor("input_393_split_cast_fp16")]; + tensor input_393_split_1_sigmoid_cast_fp16 = sigmoid(x = input_393_split_cast_fp16_1)[name = tensor("input_393_split_1_sigmoid_cast_fp16")]; + tensor input_393_cast_fp16 = mul(x = input_393_split_cast_fp16_0, y = input_393_split_1_sigmoid_cast_fp16)[name = tensor("input_393_cast_fp16")]; + tensor input_395_pad_type_0 = const()[name = tensor("input_395_pad_type_0"), val = tensor("custom")]; + tensor input_395_pad_0 = const()[name = tensor("input_395_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_395_groups_0 = const()[name = tensor("input_395_groups_0"), val = tensor(1024)]; + tensor input_395_strides_0 = const()[name = tensor("input_395_strides_0"), val = tensor([1, 1])]; + tensor input_395_dilations_0 = const()[name = tensor("input_395_dilations_0"), val = tensor([1, 1])]; + tensor const_296_to_fp16 = const()[name = tensor("const_296_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(341606720)))]; + tensor const_297_to_fp16 = const()[name = tensor("const_297_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(341625216)))]; + tensor input_397_cast_fp16 = conv(bias = const_297_to_fp16, dilations = input_395_dilations_0, groups = input_395_groups_0, pad = input_395_pad_0, pad_type = input_395_pad_type_0, strides = input_395_strides_0, weight = const_296_to_fp16, x = input_393_cast_fp16)[name = tensor("input_397_cast_fp16")]; + tensor input_399_cast_fp16 = silu(x = input_397_cast_fp16)[name = tensor("input_399_cast_fp16")]; + tensor var_5769_pad_type_0 = const()[name = tensor("op_5769_pad_type_0"), val = tensor("valid")]; + tensor var_5769_strides_0 = const()[name = tensor("op_5769_strides_0"), val = tensor([1, 1])]; + tensor var_5769_pad_0 = const()[name = tensor("op_5769_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5769_dilations_0 = const()[name = tensor("op_5769_dilations_0"), val = tensor([1, 1])]; + tensor var_5769_groups_0 = const()[name = tensor("op_5769_groups_0"), val = tensor(1)]; + tensor layers_14_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(341627328))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342413824))), name = tensor("layers_14_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5769_cast_fp16 = conv(dilations = var_5769_dilations_0, groups = var_5769_groups_0, pad = var_5769_pad_0, pad_type = var_5769_pad_type_0, strides = var_5769_strides_0, weight = layers_14_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_399_cast_fp16)[name = tensor("op_5769_cast_fp16")]; + tensor var_5775_pad_type_0 = const()[name = tensor("op_5775_pad_type_0"), val = tensor("valid")]; + tensor var_5775_strides_0 = const()[name = tensor("op_5775_strides_0"), val = tensor([1, 1])]; + tensor var_5775_pad_0 = const()[name = tensor("op_5775_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5775_dilations_0 = const()[name = tensor("op_5775_dilations_0"), val = tensor([1, 1])]; + tensor var_5775_groups_0 = const()[name = tensor("op_5775_groups_0"), val = tensor(1)]; + tensor layers_14_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342444288))), name = tensor("layers_14_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342414016))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5775_cast_fp16 = conv(dilations = var_5775_dilations_0, groups = var_5775_groups_0, pad = var_5775_pad_0, pad_type = var_5775_pad_type_0, strides = var_5775_strides_0, weight = layers_14_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_399_cast_fp16)[name = tensor("op_5775_cast_fp16")]; + tensor x_89_cast_fp16 = add(x = var_5769_cast_fp16, y = var_5775_cast_fp16)[name = tensor("x_89_cast_fp16")]; + tensor inputs_147_cast_fp16 = add(x = inputs_145_cast_fp16, y = x_89_cast_fp16)[name = tensor("inputs_147_cast_fp16")]; + tensor out_147_axes_0 = const()[name = tensor("out_147_axes_0"), val = tensor([1])]; + tensor var_5786_to_fp16 = const()[name = tensor("op_5786_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_147_cast_fp16 = layer_norm(axes = out_147_axes_0, epsilon = var_5786_to_fp16, x = inputs_147_cast_fp16)[name = tensor("out_147_cast_fp16")]; + tensor input_401_gamma_0_to_fp16 = const()[name = tensor("input_401_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342575424)))]; + tensor input_401_beta_0_to_fp16 = const()[name = tensor("input_401_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342577536)))]; + tensor input_401_epsilon_0_to_fp16 = const()[name = tensor("input_401_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_401_cast_fp16 = batch_norm(beta = input_401_beta_0_to_fp16, epsilon = input_401_epsilon_0_to_fp16, gamma = input_401_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_147_cast_fp16)[name = tensor("input_401_cast_fp16")]; + tensor var_5806_pad_type_0 = const()[name = tensor("op_5806_pad_type_0"), val = tensor("valid")]; + tensor var_5806_strides_0 = const()[name = tensor("op_5806_strides_0"), val = tensor([1, 1])]; + tensor var_5806_pad_0 = const()[name = tensor("op_5806_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5806_dilations_0 = const()[name = tensor("op_5806_dilations_0"), val = tensor([1, 1])]; + tensor var_5806_groups_0 = const()[name = tensor("op_5806_groups_0"), val = tensor(1)]; + tensor layers_14_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342579648))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345725440))), name = tensor("layers_14_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_5806_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_5806_dilations_0, groups = var_5806_groups_0, pad = var_5806_pad_0, pad_type = var_5806_pad_type_0, strides = var_5806_strides_0, weight = layers_14_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_401_cast_fp16)[name = tensor("op_5806_cast_fp16")]; + tensor var_5812_pad_type_0 = const()[name = tensor("op_5812_pad_type_0"), val = tensor("valid")]; + tensor var_5812_strides_0 = const()[name = tensor("op_5812_strides_0"), val = tensor([1, 1])]; + tensor var_5812_pad_0 = const()[name = tensor("op_5812_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5812_dilations_0 = const()[name = tensor("op_5812_dilations_0"), val = tensor([1, 1])]; + tensor var_5812_groups_0 = const()[name = tensor("op_5812_groups_0"), val = tensor(1)]; + tensor layers_14_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345876992))), name = tensor("layers_14_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345725632))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_5812_cast_fp16 = conv(dilations = var_5812_dilations_0, groups = var_5812_groups_0, pad = var_5812_pad_0, pad_type = var_5812_pad_type_0, strides = var_5812_strides_0, weight = layers_14_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_401_cast_fp16)[name = tensor("op_5812_cast_fp16")]; + tensor input_403_cast_fp16 = add(x = var_5806_cast_fp16, y = var_5812_cast_fp16)[name = tensor("input_403_cast_fp16")]; + tensor input_405_cast_fp16 = silu(x = input_403_cast_fp16)[name = tensor("input_405_cast_fp16")]; + tensor var_5823_pad_type_0 = const()[name = tensor("op_5823_pad_type_0"), val = tensor("valid")]; + tensor var_5823_strides_0 = const()[name = tensor("op_5823_strides_0"), val = tensor([1, 1])]; + tensor var_5823_pad_0 = const()[name = tensor("op_5823_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5823_dilations_0 = const()[name = tensor("op_5823_dilations_0"), val = tensor([1, 1])]; + tensor var_5823_groups_0 = const()[name = tensor("op_5823_groups_0"), val = tensor(1)]; + tensor layers_14_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(346401344))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(349547136))), name = tensor("layers_14_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_5823_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5823_dilations_0, groups = var_5823_groups_0, pad = var_5823_pad_0, pad_type = var_5823_pad_type_0, strides = var_5823_strides_0, weight = layers_14_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_405_cast_fp16)[name = tensor("op_5823_cast_fp16")]; + tensor var_5829_pad_type_0 = const()[name = tensor("op_5829_pad_type_0"), val = tensor("valid")]; + tensor var_5829_strides_0 = const()[name = tensor("op_5829_strides_0"), val = tensor([1, 1])]; + tensor var_5829_pad_0 = const()[name = tensor("op_5829_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5829_dilations_0 = const()[name = tensor("op_5829_dilations_0"), val = tensor([1, 1])]; + tensor var_5829_groups_0 = const()[name = tensor("op_5829_groups_0"), val = tensor(1)]; + tensor layers_14_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(349707328))), name = tensor("layers_14_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(349547328))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_5829_cast_fp16 = conv(dilations = var_5829_dilations_0, groups = var_5829_groups_0, pad = var_5829_pad_0, pad_type = var_5829_pad_type_0, strides = var_5829_strides_0, weight = layers_14_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_405_cast_fp16)[name = tensor("op_5829_cast_fp16")]; + tensor x_91_cast_fp16 = add(x = var_5823_cast_fp16, y = var_5829_cast_fp16)[name = tensor("x_91_cast_fp16")]; + tensor var_5831_to_fp16 = const()[name = tensor("op_5831_to_fp16"), val = tensor(0x1p-1)]; + tensor var_5832_cast_fp16 = mul(x = x_91_cast_fp16, y = var_5831_to_fp16)[name = tensor("op_5832_cast_fp16")]; + tensor inputs_149_cast_fp16 = add(x = inputs_147_cast_fp16, y = var_5832_cast_fp16)[name = tensor("inputs_149_cast_fp16")]; + tensor out_149_axes_0 = const()[name = tensor("out_149_axes_0"), val = tensor([1])]; + tensor var_5842_to_fp16 = const()[name = tensor("op_5842_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_149_cast_fp16 = layer_norm(axes = out_149_axes_0, epsilon = var_5842_to_fp16, x = inputs_149_cast_fp16)[name = tensor("out_149_cast_fp16")]; + tensor inputs_151_gamma_0_to_fp16 = const()[name = tensor("inputs_151_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(350231680)))]; + tensor inputs_151_beta_0_to_fp16 = const()[name = tensor("inputs_151_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(350233792)))]; + tensor inputs_151_epsilon_0_to_fp16 = const()[name = tensor("inputs_151_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_151_cast_fp16 = batch_norm(beta = inputs_151_beta_0_to_fp16, epsilon = inputs_151_epsilon_0_to_fp16, gamma = inputs_151_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_149_cast_fp16)[name = tensor("inputs_151_cast_fp16")]; + tensor var_5856 = const()[name = tensor("op_5856"), val = tensor(3)]; + tensor out_151_axes_0 = const()[name = tensor("out_151_axes_0"), val = tensor([1])]; + tensor var_5887_to_fp16 = const()[name = tensor("op_5887_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_151_cast_fp16 = layer_norm(axes = out_151_axes_0, epsilon = var_5887_to_fp16, x = inputs_151_cast_fp16)[name = tensor("out_151_cast_fp16")]; + tensor input_407_gamma_0_to_fp16 = const()[name = tensor("input_407_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(350235904)))]; + tensor input_407_beta_0_to_fp16 = const()[name = tensor("input_407_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(350238016)))]; + tensor input_407_epsilon_0_to_fp16 = const()[name = tensor("input_407_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_407_cast_fp16 = batch_norm(beta = input_407_beta_0_to_fp16, epsilon = input_407_epsilon_0_to_fp16, gamma = input_407_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_151_cast_fp16)[name = tensor("input_407_cast_fp16")]; + tensor var_5907_pad_type_0 = const()[name = tensor("op_5907_pad_type_0"), val = tensor("valid")]; + tensor var_5907_strides_0 = const()[name = tensor("op_5907_strides_0"), val = tensor([1, 1])]; + tensor var_5907_pad_0 = const()[name = tensor("op_5907_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5907_dilations_0 = const()[name = tensor("op_5907_dilations_0"), val = tensor([1, 1])]; + tensor var_5907_groups_0 = const()[name = tensor("op_5907_groups_0"), val = tensor(1)]; + tensor layers_15_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(350240128))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(353385920))), name = tensor("layers_15_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_5907_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_5907_dilations_0, groups = var_5907_groups_0, pad = var_5907_pad_0, pad_type = var_5907_pad_type_0, strides = var_5907_strides_0, weight = layers_15_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_407_cast_fp16)[name = tensor("op_5907_cast_fp16")]; + tensor var_5913_pad_type_0 = const()[name = tensor("op_5913_pad_type_0"), val = tensor("valid")]; + tensor var_5913_strides_0 = const()[name = tensor("op_5913_strides_0"), val = tensor([1, 1])]; + tensor var_5913_pad_0 = const()[name = tensor("op_5913_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5913_dilations_0 = const()[name = tensor("op_5913_dilations_0"), val = tensor([1, 1])]; + tensor var_5913_groups_0 = const()[name = tensor("op_5913_groups_0"), val = tensor(1)]; + tensor layers_15_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(353563712))), name = tensor("layers_15_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(353386112))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_5913_cast_fp16 = conv(dilations = var_5913_dilations_0, groups = var_5913_groups_0, pad = var_5913_pad_0, pad_type = var_5913_pad_type_0, strides = var_5913_strides_0, weight = layers_15_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_407_cast_fp16)[name = tensor("op_5913_cast_fp16")]; + tensor input_409_cast_fp16 = add(x = var_5907_cast_fp16, y = var_5913_cast_fp16)[name = tensor("input_409_cast_fp16")]; + tensor input_411_cast_fp16 = silu(x = input_409_cast_fp16)[name = tensor("input_411_cast_fp16")]; + tensor var_5924_pad_type_0 = const()[name = tensor("op_5924_pad_type_0"), val = tensor("valid")]; + tensor var_5924_strides_0 = const()[name = tensor("op_5924_strides_0"), val = tensor([1, 1])]; + tensor var_5924_pad_0 = const()[name = tensor("op_5924_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5924_dilations_0 = const()[name = tensor("op_5924_dilations_0"), val = tensor([1, 1])]; + tensor var_5924_groups_0 = const()[name = tensor("op_5924_groups_0"), val = tensor(1)]; + tensor layers_15_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(354088064))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(357233856))), name = tensor("layers_15_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_5924_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5924_dilations_0, groups = var_5924_groups_0, pad = var_5924_pad_0, pad_type = var_5924_pad_type_0, strides = var_5924_strides_0, weight = layers_15_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_411_cast_fp16)[name = tensor("op_5924_cast_fp16")]; + tensor var_5930_pad_type_0 = const()[name = tensor("op_5930_pad_type_0"), val = tensor("valid")]; + tensor var_5930_strides_0 = const()[name = tensor("op_5930_strides_0"), val = tensor([1, 1])]; + tensor var_5930_pad_0 = const()[name = tensor("op_5930_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5930_dilations_0 = const()[name = tensor("op_5930_dilations_0"), val = tensor([1, 1])]; + tensor var_5930_groups_0 = const()[name = tensor("op_5930_groups_0"), val = tensor(1)]; + tensor layers_15_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(357413056))), name = tensor("layers_15_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(357234048))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_5930_cast_fp16 = conv(dilations = var_5930_dilations_0, groups = var_5930_groups_0, pad = var_5930_pad_0, pad_type = var_5930_pad_type_0, strides = var_5930_strides_0, weight = layers_15_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_411_cast_fp16)[name = tensor("op_5930_cast_fp16")]; + tensor x_93_cast_fp16 = add(x = var_5924_cast_fp16, y = var_5930_cast_fp16)[name = tensor("x_93_cast_fp16")]; + tensor var_5932_to_fp16 = const()[name = tensor("op_5932_to_fp16"), val = tensor(0x1p-1)]; + tensor var_5933_cast_fp16 = mul(x = x_93_cast_fp16, y = var_5932_to_fp16)[name = tensor("op_5933_cast_fp16")]; + tensor inputs_153_cast_fp16 = add(x = inputs_151_cast_fp16, y = var_5933_cast_fp16)[name = tensor("inputs_153_cast_fp16")]; + tensor out_153_axes_0 = const()[name = tensor("out_153_axes_0"), val = tensor([1])]; + tensor var_5943_to_fp16 = const()[name = tensor("op_5943_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_153_cast_fp16 = layer_norm(axes = out_153_axes_0, epsilon = var_5943_to_fp16, x = inputs_153_cast_fp16)[name = tensor("out_153_cast_fp16")]; + tensor obj_63_gamma_0_to_fp16 = const()[name = tensor("obj_63_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(357937408)))]; + tensor obj_63_beta_0_to_fp16 = const()[name = tensor("obj_63_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(357939520)))]; + tensor obj_63_epsilon_0_to_fp16 = const()[name = tensor("obj_63_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_63_cast_fp16 = batch_norm(beta = obj_63_beta_0_to_fp16, epsilon = obj_63_epsilon_0_to_fp16, gamma = obj_63_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_153_cast_fp16)[name = tensor("obj_63_cast_fp16")]; + tensor var_5968_pad_type_0 = const()[name = tensor("op_5968_pad_type_0"), val = tensor("valid")]; + tensor var_5968_strides_0 = const()[name = tensor("op_5968_strides_0"), val = tensor([1, 1])]; + tensor var_5968_pad_0 = const()[name = tensor("op_5968_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5968_dilations_0 = const()[name = tensor("op_5968_dilations_0"), val = tensor([1, 1])]; + tensor var_5968_groups_0 = const()[name = tensor("op_5968_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(357941632))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(358728128))), name = tensor("layers_15_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5968_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5968_dilations_0, groups = var_5968_groups_0, pad = var_5968_pad_0, pad_type = var_5968_pad_type_0, strides = var_5968_strides_0, weight = layers_15_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_63_cast_fp16)[name = tensor("op_5968_cast_fp16")]; + tensor var_5974_pad_type_0 = const()[name = tensor("op_5974_pad_type_0"), val = tensor("valid")]; + tensor var_5974_strides_0 = const()[name = tensor("op_5974_strides_0"), val = tensor([1, 1])]; + tensor var_5974_pad_0 = const()[name = tensor("op_5974_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5974_dilations_0 = const()[name = tensor("op_5974_dilations_0"), val = tensor([1, 1])]; + tensor var_5974_groups_0 = const()[name = tensor("op_5974_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(358758592))), name = tensor("layers_15_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(358728320))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5974_cast_fp16 = conv(dilations = var_5974_dilations_0, groups = var_5974_groups_0, pad = var_5974_pad_0, pad_type = var_5974_pad_type_0, strides = var_5974_strides_0, weight = layers_15_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_63_cast_fp16)[name = tensor("op_5974_cast_fp16")]; + tensor query_61_cast_fp16 = add(x = var_5968_cast_fp16, y = var_5974_cast_fp16)[name = tensor("query_61_cast_fp16")]; + tensor var_5983_pad_type_0 = const()[name = tensor("op_5983_pad_type_0"), val = tensor("valid")]; + tensor var_5983_strides_0 = const()[name = tensor("op_5983_strides_0"), val = tensor([1, 1])]; + tensor var_5983_pad_0 = const()[name = tensor("op_5983_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5983_dilations_0 = const()[name = tensor("op_5983_dilations_0"), val = tensor([1, 1])]; + tensor var_5983_groups_0 = const()[name = tensor("op_5983_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(358889728))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(359676224))), name = tensor("layers_15_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5983_cast_fp16 = conv(dilations = var_5983_dilations_0, groups = var_5983_groups_0, pad = var_5983_pad_0, pad_type = var_5983_pad_type_0, strides = var_5983_strides_0, weight = layers_15_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_63_cast_fp16)[name = tensor("op_5983_cast_fp16")]; + tensor var_5989_pad_type_0 = const()[name = tensor("op_5989_pad_type_0"), val = tensor("valid")]; + tensor var_5989_strides_0 = const()[name = tensor("op_5989_strides_0"), val = tensor([1, 1])]; + tensor var_5989_pad_0 = const()[name = tensor("op_5989_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5989_dilations_0 = const()[name = tensor("op_5989_dilations_0"), val = tensor([1, 1])]; + tensor var_5989_groups_0 = const()[name = tensor("op_5989_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(359713216))), name = tensor("layers_15_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(359676416))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5989_cast_fp16 = conv(dilations = var_5989_dilations_0, groups = var_5989_groups_0, pad = var_5989_pad_0, pad_type = var_5989_pad_type_0, strides = var_5989_strides_0, weight = layers_15_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_63_cast_fp16)[name = tensor("op_5989_cast_fp16")]; + tensor key_31_cast_fp16 = add(x = var_5983_cast_fp16, y = var_5989_cast_fp16)[name = tensor("key_31_cast_fp16")]; + tensor var_5999_pad_type_0 = const()[name = tensor("op_5999_pad_type_0"), val = tensor("valid")]; + tensor var_5999_strides_0 = const()[name = tensor("op_5999_strides_0"), val = tensor([1, 1])]; + tensor var_5999_pad_0 = const()[name = tensor("op_5999_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5999_dilations_0 = const()[name = tensor("op_5999_dilations_0"), val = tensor([1, 1])]; + tensor var_5999_groups_0 = const()[name = tensor("op_5999_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(359844352))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(360630848))), name = tensor("layers_15_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_5999_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5999_dilations_0, groups = var_5999_groups_0, pad = var_5999_pad_0, pad_type = var_5999_pad_type_0, strides = var_5999_strides_0, weight = layers_15_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_63_cast_fp16)[name = tensor("op_5999_cast_fp16")]; + tensor var_6005_pad_type_0 = const()[name = tensor("op_6005_pad_type_0"), val = tensor("valid")]; + tensor var_6005_strides_0 = const()[name = tensor("op_6005_strides_0"), val = tensor([1, 1])]; + tensor var_6005_pad_0 = const()[name = tensor("op_6005_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6005_dilations_0 = const()[name = tensor("op_6005_dilations_0"), val = tensor([1, 1])]; + tensor var_6005_groups_0 = const()[name = tensor("op_6005_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(360661824))), name = tensor("layers_15_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(360631040))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6005_cast_fp16 = conv(dilations = var_6005_dilations_0, groups = var_6005_groups_0, pad = var_6005_pad_0, pad_type = var_6005_pad_type_0, strides = var_6005_strides_0, weight = layers_15_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_63_cast_fp16)[name = tensor("op_6005_cast_fp16")]; + tensor value_31_cast_fp16 = add(x = var_5999_cast_fp16, y = var_6005_cast_fp16)[name = tensor("value_31_cast_fp16")]; + tensor var_6008_to_fp16 = const()[name = tensor("op_6008_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(360792960)))]; + tensor query_63_cast_fp16 = add(x = query_61_cast_fp16, y = var_6008_to_fp16)[name = tensor("query_63_cast_fp16")]; + tensor var_6011_to_fp16 = const()[name = tensor("op_6011_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(360795072)))]; + tensor q_with_bias_v_31_cast_fp16 = add(x = query_61_cast_fp16, y = var_6011_to_fp16)[name = tensor("q_with_bias_v_31_cast_fp16")]; + tensor var_6021_pad_type_0 = const()[name = tensor("op_6021_pad_type_0"), val = tensor("valid")]; + tensor var_6021_strides_0 = const()[name = tensor("op_6021_strides_0"), val = tensor([1, 1])]; + tensor var_6021_pad_0 = const()[name = tensor("op_6021_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6021_dilations_0 = const()[name = tensor("op_6021_dilations_0"), val = tensor([1, 1])]; + tensor var_6021_groups_0 = const()[name = tensor("op_6021_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(360797184))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(361583680))), name = tensor("layers_15_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6021_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6021_dilations_0, groups = var_6021_groups_0, pad = var_6021_pad_0, pad_type = var_6021_pad_type_0, strides = var_6021_strides_0, weight = layers_15_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_6021_cast_fp16")]; + tensor var_6027_pad_type_0 = const()[name = tensor("op_6027_pad_type_0"), val = tensor("valid")]; + tensor var_6027_strides_0 = const()[name = tensor("op_6027_strides_0"), val = tensor([1, 1])]; + tensor var_6027_pad_0 = const()[name = tensor("op_6027_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6027_dilations_0 = const()[name = tensor("op_6027_dilations_0"), val = tensor([1, 1])]; + tensor var_6027_groups_0 = const()[name = tensor("op_6027_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(361640448))), name = tensor("layers_15_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(361583872))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6027_cast_fp16 = conv(dilations = var_6027_dilations_0, groups = var_6027_groups_0, pad = var_6027_pad_0, pad_type = var_6027_pad_type_0, strides = var_6027_strides_0, weight = layers_15_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_6027_cast_fp16")]; + tensor p_31_cast_fp16 = add(x = var_6021_cast_fp16, y = var_6027_cast_fp16)[name = tensor("p_31_cast_fp16")]; + tensor var_6031 = const()[name = tensor("op_6031"), val = tensor([1, 8, 128, 188])]; + tensor var_6032_cast_fp16 = reshape(shape = var_6031, x = q_with_bias_v_31_cast_fp16)[name = tensor("op_6032_cast_fp16")]; + tensor var_6033 = const()[name = tensor("op_6033"), val = tensor([1, 8, 128, -1])]; + tensor var_6034_cast_fp16 = reshape(shape = var_6033, x = p_31_cast_fp16)[name = tensor("op_6034_cast_fp16")]; + tensor matrix_bd_121_transpose_x_0 = const()[name = tensor("matrix_bd_121_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_121_transpose_y_0 = const()[name = tensor("matrix_bd_121_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_121_cast_fp16 = matmul(transpose_x = matrix_bd_121_transpose_x_0, transpose_y = matrix_bd_121_transpose_y_0, x = var_6032_cast_fp16, y = var_6034_cast_fp16)[name = tensor("matrix_bd_121_cast_fp16")]; + tensor matrix_bd_123_pad_0 = const()[name = tensor("matrix_bd_123_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_123_mode_0 = const()[name = tensor("matrix_bd_123_mode_0"), val = tensor("constant")]; + tensor const_175_to_fp16 = const()[name = tensor("const_175_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_123_cast_fp16 = pad(constant_val = const_175_to_fp16, mode = matrix_bd_123_mode_0, pad = matrix_bd_123_pad_0, x = matrix_bd_121_cast_fp16)[name = tensor("matrix_bd_123_cast_fp16")]; + tensor var_6043 = const()[name = tensor("op_6043"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_125_cast_fp16 = reshape(shape = var_6043, x = matrix_bd_123_cast_fp16)[name = tensor("matrix_bd_125_cast_fp16")]; + tensor var_6047_begin_0 = const()[name = tensor("op_6047_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_6047_end_0 = const()[name = tensor("op_6047_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_6047_end_mask_0 = const()[name = tensor("op_6047_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_6047_cast_fp16 = slice_by_index(begin = var_6047_begin_0, end = var_6047_end_0, end_mask = var_6047_end_mask_0, x = matrix_bd_125_cast_fp16)[name = tensor("op_6047_cast_fp16")]; + tensor var_6048 = const()[name = tensor("op_6048"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_127_cast_fp16 = reshape(shape = var_6048, x = var_6047_cast_fp16)[name = tensor("matrix_bd_127_cast_fp16")]; + tensor var_6053_begin_0 = const()[name = tensor("op_6053_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6053_end_0 = const()[name = tensor("op_6053_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_6053_end_mask_0 = const()[name = tensor("op_6053_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6053_cast_fp16 = slice_by_index(begin = var_6053_begin_0, end = var_6053_end_0, end_mask = var_6053_end_mask_0, x = matrix_bd_127_cast_fp16)[name = tensor("op_6053_cast_fp16")]; + tensor var_6054_to_fp16 = const()[name = tensor("op_6054_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_31_cast_fp16 = mul(x = var_6053_cast_fp16, y = var_6054_to_fp16)[name = tensor("qk_mask_31_cast_fp16")]; + tensor var_6058 = const()[name = tensor("op_6058"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_31_cast_fp16 = reshape(shape = var_6058, x = query_63_cast_fp16)[name = tensor("mh_q_31_cast_fp16")]; + tensor var_6060_to_fp16 = const()[name = tensor("op_6060_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_6061_cast_fp16 = mul(x = mh_q_31_cast_fp16, y = var_6060_to_fp16)[name = tensor("op_6061_cast_fp16")]; + tensor var_6064 = const()[name = tensor("op_6064"), val = tensor([1, 8, 128, 188])]; + tensor var_6065_cast_fp16 = reshape(shape = var_6064, x = key_31_cast_fp16)[name = tensor("op_6065_cast_fp16")]; + tensor mh_w_61_transpose_x_0 = const()[name = tensor("mh_w_61_transpose_x_0"), val = tensor(true)]; + tensor mh_w_61_transpose_y_0 = const()[name = tensor("mh_w_61_transpose_y_0"), val = tensor(false)]; + tensor mh_w_61_cast_fp16 = matmul(transpose_x = mh_w_61_transpose_x_0, transpose_y = mh_w_61_transpose_y_0, x = var_6061_cast_fp16, y = var_6065_cast_fp16)[name = tensor("mh_w_61_cast_fp16")]; + tensor mh_w_63_cast_fp16 = add(x = mh_w_61_cast_fp16, y = qk_mask_31_cast_fp16)[name = tensor("mh_w_63_cast_fp16")]; + tensor var_6069_cast_fp16 = softmax(axis = var_5856, x = mh_w_63_cast_fp16)[name = tensor("op_6069_cast_fp16")]; + tensor var_6070 = const()[name = tensor("op_6070"), val = tensor([1, 8, 128, 188])]; + tensor var_6071_cast_fp16 = reshape(shape = var_6070, x = value_31_cast_fp16)[name = tensor("op_6071_cast_fp16")]; + tensor attn_31_transpose_x_0 = const()[name = tensor("attn_31_transpose_x_0"), val = tensor(false)]; + tensor attn_31_transpose_y_0 = const()[name = tensor("attn_31_transpose_y_0"), val = tensor(true)]; + tensor attn_31_cast_fp16 = matmul(transpose_x = attn_31_transpose_x_0, transpose_y = attn_31_transpose_y_0, x = var_6071_cast_fp16, y = var_6069_cast_fp16)[name = tensor("attn_31_cast_fp16")]; + tensor var_6074 = const()[name = tensor("op_6074"), val = tensor([1, 1024, 1, 188])]; + tensor input_413_cast_fp16 = reshape(shape = var_6074, x = attn_31_cast_fp16)[name = tensor("input_413_cast_fp16")]; + tensor var_6084_pad_type_0 = const()[name = tensor("op_6084_pad_type_0"), val = tensor("valid")]; + tensor var_6084_strides_0 = const()[name = tensor("op_6084_strides_0"), val = tensor([1, 1])]; + tensor var_6084_pad_0 = const()[name = tensor("op_6084_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6084_dilations_0 = const()[name = tensor("op_6084_dilations_0"), val = tensor([1, 1])]; + tensor var_6084_groups_0 = const()[name = tensor("op_6084_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(361771584))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362558080))), name = tensor("layers_15_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6084_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6084_dilations_0, groups = var_6084_groups_0, pad = var_6084_pad_0, pad_type = var_6084_pad_type_0, strides = var_6084_strides_0, weight = layers_15_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_413_cast_fp16)[name = tensor("op_6084_cast_fp16")]; + tensor var_6090_pad_type_0 = const()[name = tensor("op_6090_pad_type_0"), val = tensor("valid")]; + tensor var_6090_strides_0 = const()[name = tensor("op_6090_strides_0"), val = tensor([1, 1])]; + tensor var_6090_pad_0 = const()[name = tensor("op_6090_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6090_dilations_0 = const()[name = tensor("op_6090_dilations_0"), val = tensor([1, 1])]; + tensor var_6090_groups_0 = const()[name = tensor("op_6090_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362588864))), name = tensor("layers_15_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362558272))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6090_cast_fp16 = conv(dilations = var_6090_dilations_0, groups = var_6090_groups_0, pad = var_6090_pad_0, pad_type = var_6090_pad_type_0, strides = var_6090_strides_0, weight = layers_15_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_413_cast_fp16)[name = tensor("op_6090_cast_fp16")]; + tensor obj_65_cast_fp16 = add(x = var_6084_cast_fp16, y = var_6090_cast_fp16)[name = tensor("obj_65_cast_fp16")]; + tensor inputs_155_cast_fp16 = add(x = inputs_153_cast_fp16, y = obj_65_cast_fp16)[name = tensor("inputs_155_cast_fp16")]; + tensor out_155_axes_0 = const()[name = tensor("out_155_axes_0"), val = tensor([1])]; + tensor var_6101_to_fp16 = const()[name = tensor("op_6101_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_155_cast_fp16 = layer_norm(axes = out_155_axes_0, epsilon = var_6101_to_fp16, x = inputs_155_cast_fp16)[name = tensor("out_155_cast_fp16")]; + tensor input_415_gamma_0_to_fp16 = const()[name = tensor("input_415_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362720000)))]; + tensor input_415_beta_0_to_fp16 = const()[name = tensor("input_415_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362722112)))]; + tensor input_415_epsilon_0_to_fp16 = const()[name = tensor("input_415_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_415_cast_fp16 = batch_norm(beta = input_415_beta_0_to_fp16, epsilon = input_415_epsilon_0_to_fp16, gamma = input_415_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_155_cast_fp16)[name = tensor("input_415_cast_fp16")]; + tensor var_6122_pad_type_0 = const()[name = tensor("op_6122_pad_type_0"), val = tensor("valid")]; + tensor var_6122_strides_0 = const()[name = tensor("op_6122_strides_0"), val = tensor([1, 1])]; + tensor var_6122_pad_0 = const()[name = tensor("op_6122_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6122_dilations_0 = const()[name = tensor("op_6122_dilations_0"), val = tensor([1, 1])]; + tensor var_6122_groups_0 = const()[name = tensor("op_6122_groups_0"), val = tensor(1)]; + tensor layers_15_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362724224))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364297152))), name = tensor("layers_15_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_6122_cast_fp16 = conv(dilations = var_6122_dilations_0, groups = var_6122_groups_0, pad = var_6122_pad_0, pad_type = var_6122_pad_type_0, strides = var_6122_strides_0, weight = layers_15_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_415_cast_fp16)[name = tensor("op_6122_cast_fp16")]; + tensor var_6128_pad_type_0 = const()[name = tensor("op_6128_pad_type_0"), val = tensor("valid")]; + tensor var_6128_strides_0 = const()[name = tensor("op_6128_strides_0"), val = tensor([1, 1])]; + tensor var_6128_pad_0 = const()[name = tensor("op_6128_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6128_dilations_0 = const()[name = tensor("op_6128_dilations_0"), val = tensor([1, 1])]; + tensor var_6128_groups_0 = const()[name = tensor("op_6128_groups_0"), val = tensor(1)]; + tensor layers_15_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364360128))), name = tensor("layers_15_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364297344))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_6128_cast_fp16 = conv(dilations = var_6128_dilations_0, groups = var_6128_groups_0, pad = var_6128_pad_0, pad_type = var_6128_pad_type_0, strides = var_6128_strides_0, weight = layers_15_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_415_cast_fp16)[name = tensor("op_6128_cast_fp16")]; + tensor input_417_cast_fp16 = add(x = var_6122_cast_fp16, y = var_6128_cast_fp16)[name = tensor("input_417_cast_fp16")]; + tensor input_419_split_num_splits_0 = const()[name = tensor("input_419_split_num_splits_0"), val = tensor(2)]; + tensor input_419_split_axis_0 = const()[name = tensor("input_419_split_axis_0"), val = tensor(1)]; + tensor input_419_split_cast_fp16_0, tensor input_419_split_cast_fp16_1 = split(axis = input_419_split_axis_0, num_splits = input_419_split_num_splits_0, x = input_417_cast_fp16)[name = tensor("input_419_split_cast_fp16")]; + tensor input_419_split_1_sigmoid_cast_fp16 = sigmoid(x = input_419_split_cast_fp16_1)[name = tensor("input_419_split_1_sigmoid_cast_fp16")]; + tensor input_419_cast_fp16 = mul(x = input_419_split_cast_fp16_0, y = input_419_split_1_sigmoid_cast_fp16)[name = tensor("input_419_cast_fp16")]; + tensor input_421_pad_type_0 = const()[name = tensor("input_421_pad_type_0"), val = tensor("custom")]; + tensor input_421_pad_0 = const()[name = tensor("input_421_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_421_groups_0 = const()[name = tensor("input_421_groups_0"), val = tensor(1024)]; + tensor input_421_strides_0 = const()[name = tensor("input_421_strides_0"), val = tensor([1, 1])]; + tensor input_421_dilations_0 = const()[name = tensor("input_421_dilations_0"), val = tensor([1, 1])]; + tensor const_298_to_fp16 = const()[name = tensor("const_298_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364622336)))]; + tensor const_299_to_fp16 = const()[name = tensor("const_299_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364640832)))]; + tensor input_423_cast_fp16 = conv(bias = const_299_to_fp16, dilations = input_421_dilations_0, groups = input_421_groups_0, pad = input_421_pad_0, pad_type = input_421_pad_type_0, strides = input_421_strides_0, weight = const_298_to_fp16, x = input_419_cast_fp16)[name = tensor("input_423_cast_fp16")]; + tensor input_425_cast_fp16 = silu(x = input_423_cast_fp16)[name = tensor("input_425_cast_fp16")]; + tensor var_6150_pad_type_0 = const()[name = tensor("op_6150_pad_type_0"), val = tensor("valid")]; + tensor var_6150_strides_0 = const()[name = tensor("op_6150_strides_0"), val = tensor([1, 1])]; + tensor var_6150_pad_0 = const()[name = tensor("op_6150_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6150_dilations_0 = const()[name = tensor("op_6150_dilations_0"), val = tensor([1, 1])]; + tensor var_6150_groups_0 = const()[name = tensor("op_6150_groups_0"), val = tensor(1)]; + tensor layers_15_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364642944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(365429440))), name = tensor("layers_15_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6150_cast_fp16 = conv(dilations = var_6150_dilations_0, groups = var_6150_groups_0, pad = var_6150_pad_0, pad_type = var_6150_pad_type_0, strides = var_6150_strides_0, weight = layers_15_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_425_cast_fp16)[name = tensor("op_6150_cast_fp16")]; + tensor var_6156_pad_type_0 = const()[name = tensor("op_6156_pad_type_0"), val = tensor("valid")]; + tensor var_6156_strides_0 = const()[name = tensor("op_6156_strides_0"), val = tensor([1, 1])]; + tensor var_6156_pad_0 = const()[name = tensor("op_6156_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6156_dilations_0 = const()[name = tensor("op_6156_dilations_0"), val = tensor([1, 1])]; + tensor var_6156_groups_0 = const()[name = tensor("op_6156_groups_0"), val = tensor(1)]; + tensor layers_15_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(365460736))), name = tensor("layers_15_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(365429632))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6156_cast_fp16 = conv(dilations = var_6156_dilations_0, groups = var_6156_groups_0, pad = var_6156_pad_0, pad_type = var_6156_pad_type_0, strides = var_6156_strides_0, weight = layers_15_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_425_cast_fp16)[name = tensor("op_6156_cast_fp16")]; + tensor x_95_cast_fp16 = add(x = var_6150_cast_fp16, y = var_6156_cast_fp16)[name = tensor("x_95_cast_fp16")]; + tensor inputs_157_cast_fp16 = add(x = inputs_155_cast_fp16, y = x_95_cast_fp16)[name = tensor("inputs_157_cast_fp16")]; + tensor out_157_axes_0 = const()[name = tensor("out_157_axes_0"), val = tensor([1])]; + tensor var_6167_to_fp16 = const()[name = tensor("op_6167_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_157_cast_fp16 = layer_norm(axes = out_157_axes_0, epsilon = var_6167_to_fp16, x = inputs_157_cast_fp16)[name = tensor("out_157_cast_fp16")]; + tensor input_427_gamma_0_to_fp16 = const()[name = tensor("input_427_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(365591872)))]; + tensor input_427_beta_0_to_fp16 = const()[name = tensor("input_427_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(365593984)))]; + tensor input_427_epsilon_0_to_fp16 = const()[name = tensor("input_427_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_427_cast_fp16 = batch_norm(beta = input_427_beta_0_to_fp16, epsilon = input_427_epsilon_0_to_fp16, gamma = input_427_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_157_cast_fp16)[name = tensor("input_427_cast_fp16")]; + tensor var_6187_pad_type_0 = const()[name = tensor("op_6187_pad_type_0"), val = tensor("valid")]; + tensor var_6187_strides_0 = const()[name = tensor("op_6187_strides_0"), val = tensor([1, 1])]; + tensor var_6187_pad_0 = const()[name = tensor("op_6187_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6187_dilations_0 = const()[name = tensor("op_6187_dilations_0"), val = tensor([1, 1])]; + tensor var_6187_groups_0 = const()[name = tensor("op_6187_groups_0"), val = tensor(1)]; + tensor layers_15_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(365596096))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368741888))), name = tensor("layers_15_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_6187_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_6187_dilations_0, groups = var_6187_groups_0, pad = var_6187_pad_0, pad_type = var_6187_pad_type_0, strides = var_6187_strides_0, weight = layers_15_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_427_cast_fp16)[name = tensor("op_6187_cast_fp16")]; + tensor var_6193_pad_type_0 = const()[name = tensor("op_6193_pad_type_0"), val = tensor("valid")]; + tensor var_6193_strides_0 = const()[name = tensor("op_6193_strides_0"), val = tensor([1, 1])]; + tensor var_6193_pad_0 = const()[name = tensor("op_6193_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6193_dilations_0 = const()[name = tensor("op_6193_dilations_0"), val = tensor([1, 1])]; + tensor var_6193_groups_0 = const()[name = tensor("op_6193_groups_0"), val = tensor(1)]; + tensor layers_15_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368897472))), name = tensor("layers_15_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368742080))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_6193_cast_fp16 = conv(dilations = var_6193_dilations_0, groups = var_6193_groups_0, pad = var_6193_pad_0, pad_type = var_6193_pad_type_0, strides = var_6193_strides_0, weight = layers_15_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_427_cast_fp16)[name = tensor("op_6193_cast_fp16")]; + tensor input_429_cast_fp16 = add(x = var_6187_cast_fp16, y = var_6193_cast_fp16)[name = tensor("input_429_cast_fp16")]; + tensor input_431_cast_fp16 = silu(x = input_429_cast_fp16)[name = tensor("input_431_cast_fp16")]; + tensor var_6204_pad_type_0 = const()[name = tensor("op_6204_pad_type_0"), val = tensor("valid")]; + tensor var_6204_strides_0 = const()[name = tensor("op_6204_strides_0"), val = tensor([1, 1])]; + tensor var_6204_pad_0 = const()[name = tensor("op_6204_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6204_dilations_0 = const()[name = tensor("op_6204_dilations_0"), val = tensor([1, 1])]; + tensor var_6204_groups_0 = const()[name = tensor("op_6204_groups_0"), val = tensor(1)]; + tensor layers_15_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(369421824))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(372567616))), name = tensor("layers_15_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_6204_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6204_dilations_0, groups = var_6204_groups_0, pad = var_6204_pad_0, pad_type = var_6204_pad_type_0, strides = var_6204_strides_0, weight = layers_15_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_431_cast_fp16)[name = tensor("op_6204_cast_fp16")]; + tensor var_6210_pad_type_0 = const()[name = tensor("op_6210_pad_type_0"), val = tensor("valid")]; + tensor var_6210_strides_0 = const()[name = tensor("op_6210_strides_0"), val = tensor([1, 1])]; + tensor var_6210_pad_0 = const()[name = tensor("op_6210_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6210_dilations_0 = const()[name = tensor("op_6210_dilations_0"), val = tensor([1, 1])]; + tensor var_6210_groups_0 = const()[name = tensor("op_6210_groups_0"), val = tensor(1)]; + tensor layers_15_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(372731648))), name = tensor("layers_15_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(372567808))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_6210_cast_fp16 = conv(dilations = var_6210_dilations_0, groups = var_6210_groups_0, pad = var_6210_pad_0, pad_type = var_6210_pad_type_0, strides = var_6210_strides_0, weight = layers_15_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_431_cast_fp16)[name = tensor("op_6210_cast_fp16")]; + tensor x_97_cast_fp16 = add(x = var_6204_cast_fp16, y = var_6210_cast_fp16)[name = tensor("x_97_cast_fp16")]; + tensor var_6212_to_fp16 = const()[name = tensor("op_6212_to_fp16"), val = tensor(0x1p-1)]; + tensor var_6213_cast_fp16 = mul(x = x_97_cast_fp16, y = var_6212_to_fp16)[name = tensor("op_6213_cast_fp16")]; + tensor inputs_159_cast_fp16 = add(x = inputs_157_cast_fp16, y = var_6213_cast_fp16)[name = tensor("inputs_159_cast_fp16")]; + tensor out_159_axes_0 = const()[name = tensor("out_159_axes_0"), val = tensor([1])]; + tensor var_6223_to_fp16 = const()[name = tensor("op_6223_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_159_cast_fp16 = layer_norm(axes = out_159_axes_0, epsilon = var_6223_to_fp16, x = inputs_159_cast_fp16)[name = tensor("out_159_cast_fp16")]; + tensor inputs_161_gamma_0_to_fp16 = const()[name = tensor("inputs_161_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(373256000)))]; + tensor inputs_161_beta_0_to_fp16 = const()[name = tensor("inputs_161_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(373258112)))]; + tensor inputs_161_epsilon_0_to_fp16 = const()[name = tensor("inputs_161_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_161_cast_fp16 = batch_norm(beta = inputs_161_beta_0_to_fp16, epsilon = inputs_161_epsilon_0_to_fp16, gamma = inputs_161_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_159_cast_fp16)[name = tensor("inputs_161_cast_fp16")]; + tensor var_6237 = const()[name = tensor("op_6237"), val = tensor(3)]; + tensor out_161_axes_0 = const()[name = tensor("out_161_axes_0"), val = tensor([1])]; + tensor var_6268_to_fp16 = const()[name = tensor("op_6268_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_161_cast_fp16 = layer_norm(axes = out_161_axes_0, epsilon = var_6268_to_fp16, x = inputs_161_cast_fp16)[name = tensor("out_161_cast_fp16")]; + tensor input_433_gamma_0_to_fp16 = const()[name = tensor("input_433_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(373260224)))]; + tensor input_433_beta_0_to_fp16 = const()[name = tensor("input_433_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(373262336)))]; + tensor input_433_epsilon_0_to_fp16 = const()[name = tensor("input_433_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_433_cast_fp16 = batch_norm(beta = input_433_beta_0_to_fp16, epsilon = input_433_epsilon_0_to_fp16, gamma = input_433_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_161_cast_fp16)[name = tensor("input_433_cast_fp16")]; + tensor var_6288_pad_type_0 = const()[name = tensor("op_6288_pad_type_0"), val = tensor("valid")]; + tensor var_6288_strides_0 = const()[name = tensor("op_6288_strides_0"), val = tensor([1, 1])]; + tensor var_6288_pad_0 = const()[name = tensor("op_6288_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6288_dilations_0 = const()[name = tensor("op_6288_dilations_0"), val = tensor([1, 1])]; + tensor var_6288_groups_0 = const()[name = tensor("op_6288_groups_0"), val = tensor(1)]; + tensor layers_16_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(373264448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(376410240))), name = tensor("layers_16_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_6288_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_6288_dilations_0, groups = var_6288_groups_0, pad = var_6288_pad_0, pad_type = var_6288_pad_type_0, strides = var_6288_strides_0, weight = layers_16_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_433_cast_fp16)[name = tensor("op_6288_cast_fp16")]; + tensor var_6294_pad_type_0 = const()[name = tensor("op_6294_pad_type_0"), val = tensor("valid")]; + tensor var_6294_strides_0 = const()[name = tensor("op_6294_strides_0"), val = tensor([1, 1])]; + tensor var_6294_pad_0 = const()[name = tensor("op_6294_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6294_dilations_0 = const()[name = tensor("op_6294_dilations_0"), val = tensor([1, 1])]; + tensor var_6294_groups_0 = const()[name = tensor("op_6294_groups_0"), val = tensor(1)]; + tensor layers_16_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(376583552))), name = tensor("layers_16_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(376410432))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_6294_cast_fp16 = conv(dilations = var_6294_dilations_0, groups = var_6294_groups_0, pad = var_6294_pad_0, pad_type = var_6294_pad_type_0, strides = var_6294_strides_0, weight = layers_16_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_433_cast_fp16)[name = tensor("op_6294_cast_fp16")]; + tensor input_435_cast_fp16 = add(x = var_6288_cast_fp16, y = var_6294_cast_fp16)[name = tensor("input_435_cast_fp16")]; + tensor input_437_cast_fp16 = silu(x = input_435_cast_fp16)[name = tensor("input_437_cast_fp16")]; + tensor var_6305_pad_type_0 = const()[name = tensor("op_6305_pad_type_0"), val = tensor("valid")]; + tensor var_6305_strides_0 = const()[name = tensor("op_6305_strides_0"), val = tensor([1, 1])]; + tensor var_6305_pad_0 = const()[name = tensor("op_6305_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6305_dilations_0 = const()[name = tensor("op_6305_dilations_0"), val = tensor([1, 1])]; + tensor var_6305_groups_0 = const()[name = tensor("op_6305_groups_0"), val = tensor(1)]; + tensor layers_16_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(377107904))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(380253696))), name = tensor("layers_16_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_6305_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6305_dilations_0, groups = var_6305_groups_0, pad = var_6305_pad_0, pad_type = var_6305_pad_type_0, strides = var_6305_strides_0, weight = layers_16_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_437_cast_fp16)[name = tensor("op_6305_cast_fp16")]; + tensor var_6311_pad_type_0 = const()[name = tensor("op_6311_pad_type_0"), val = tensor("valid")]; + tensor var_6311_strides_0 = const()[name = tensor("op_6311_strides_0"), val = tensor([1, 1])]; + tensor var_6311_pad_0 = const()[name = tensor("op_6311_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6311_dilations_0 = const()[name = tensor("op_6311_dilations_0"), val = tensor([1, 1])]; + tensor var_6311_groups_0 = const()[name = tensor("op_6311_groups_0"), val = tensor(1)]; + tensor layers_16_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(380430336))), name = tensor("layers_16_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(380253888))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_6311_cast_fp16 = conv(dilations = var_6311_dilations_0, groups = var_6311_groups_0, pad = var_6311_pad_0, pad_type = var_6311_pad_type_0, strides = var_6311_strides_0, weight = layers_16_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_437_cast_fp16)[name = tensor("op_6311_cast_fp16")]; + tensor x_99_cast_fp16 = add(x = var_6305_cast_fp16, y = var_6311_cast_fp16)[name = tensor("x_99_cast_fp16")]; + tensor var_6313_to_fp16 = const()[name = tensor("op_6313_to_fp16"), val = tensor(0x1p-1)]; + tensor var_6314_cast_fp16 = mul(x = x_99_cast_fp16, y = var_6313_to_fp16)[name = tensor("op_6314_cast_fp16")]; + tensor inputs_163_cast_fp16 = add(x = inputs_161_cast_fp16, y = var_6314_cast_fp16)[name = tensor("inputs_163_cast_fp16")]; + tensor out_163_axes_0 = const()[name = tensor("out_163_axes_0"), val = tensor([1])]; + tensor var_6324_to_fp16 = const()[name = tensor("op_6324_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_163_cast_fp16 = layer_norm(axes = out_163_axes_0, epsilon = var_6324_to_fp16, x = inputs_163_cast_fp16)[name = tensor("out_163_cast_fp16")]; + tensor obj_67_gamma_0_to_fp16 = const()[name = tensor("obj_67_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(380954688)))]; + tensor obj_67_beta_0_to_fp16 = const()[name = tensor("obj_67_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(380956800)))]; + tensor obj_67_epsilon_0_to_fp16 = const()[name = tensor("obj_67_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_67_cast_fp16 = batch_norm(beta = obj_67_beta_0_to_fp16, epsilon = obj_67_epsilon_0_to_fp16, gamma = obj_67_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_163_cast_fp16)[name = tensor("obj_67_cast_fp16")]; + tensor var_6349_pad_type_0 = const()[name = tensor("op_6349_pad_type_0"), val = tensor("valid")]; + tensor var_6349_strides_0 = const()[name = tensor("op_6349_strides_0"), val = tensor([1, 1])]; + tensor var_6349_pad_0 = const()[name = tensor("op_6349_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6349_dilations_0 = const()[name = tensor("op_6349_dilations_0"), val = tensor([1, 1])]; + tensor var_6349_groups_0 = const()[name = tensor("op_6349_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(380958912))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381745408))), name = tensor("layers_16_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6349_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6349_dilations_0, groups = var_6349_groups_0, pad = var_6349_pad_0, pad_type = var_6349_pad_type_0, strides = var_6349_strides_0, weight = layers_16_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_67_cast_fp16)[name = tensor("op_6349_cast_fp16")]; + tensor var_6355_pad_type_0 = const()[name = tensor("op_6355_pad_type_0"), val = tensor("valid")]; + tensor var_6355_strides_0 = const()[name = tensor("op_6355_strides_0"), val = tensor([1, 1])]; + tensor var_6355_pad_0 = const()[name = tensor("op_6355_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6355_dilations_0 = const()[name = tensor("op_6355_dilations_0"), val = tensor([1, 1])]; + tensor var_6355_groups_0 = const()[name = tensor("op_6355_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381775744))), name = tensor("layers_16_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381745600))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6355_cast_fp16 = conv(dilations = var_6355_dilations_0, groups = var_6355_groups_0, pad = var_6355_pad_0, pad_type = var_6355_pad_type_0, strides = var_6355_strides_0, weight = layers_16_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_67_cast_fp16)[name = tensor("op_6355_cast_fp16")]; + tensor query_65_cast_fp16 = add(x = var_6349_cast_fp16, y = var_6355_cast_fp16)[name = tensor("query_65_cast_fp16")]; + tensor var_6364_pad_type_0 = const()[name = tensor("op_6364_pad_type_0"), val = tensor("valid")]; + tensor var_6364_strides_0 = const()[name = tensor("op_6364_strides_0"), val = tensor([1, 1])]; + tensor var_6364_pad_0 = const()[name = tensor("op_6364_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6364_dilations_0 = const()[name = tensor("op_6364_dilations_0"), val = tensor([1, 1])]; + tensor var_6364_groups_0 = const()[name = tensor("op_6364_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381906880))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(382693376))), name = tensor("layers_16_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6364_cast_fp16 = conv(dilations = var_6364_dilations_0, groups = var_6364_groups_0, pad = var_6364_pad_0, pad_type = var_6364_pad_type_0, strides = var_6364_strides_0, weight = layers_16_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_67_cast_fp16)[name = tensor("op_6364_cast_fp16")]; + tensor var_6370_pad_type_0 = const()[name = tensor("op_6370_pad_type_0"), val = tensor("valid")]; + tensor var_6370_strides_0 = const()[name = tensor("op_6370_strides_0"), val = tensor([1, 1])]; + tensor var_6370_pad_0 = const()[name = tensor("op_6370_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6370_dilations_0 = const()[name = tensor("op_6370_dilations_0"), val = tensor([1, 1])]; + tensor var_6370_groups_0 = const()[name = tensor("op_6370_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(382735168))), name = tensor("layers_16_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(382693568))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6370_cast_fp16 = conv(dilations = var_6370_dilations_0, groups = var_6370_groups_0, pad = var_6370_pad_0, pad_type = var_6370_pad_type_0, strides = var_6370_strides_0, weight = layers_16_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_67_cast_fp16)[name = tensor("op_6370_cast_fp16")]; + tensor key_33_cast_fp16 = add(x = var_6364_cast_fp16, y = var_6370_cast_fp16)[name = tensor("key_33_cast_fp16")]; + tensor var_6380_pad_type_0 = const()[name = tensor("op_6380_pad_type_0"), val = tensor("valid")]; + tensor var_6380_strides_0 = const()[name = tensor("op_6380_strides_0"), val = tensor([1, 1])]; + tensor var_6380_pad_0 = const()[name = tensor("op_6380_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6380_dilations_0 = const()[name = tensor("op_6380_dilations_0"), val = tensor([1, 1])]; + tensor var_6380_groups_0 = const()[name = tensor("op_6380_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(382866304))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(383652800))), name = tensor("layers_16_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6380_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6380_dilations_0, groups = var_6380_groups_0, pad = var_6380_pad_0, pad_type = var_6380_pad_type_0, strides = var_6380_strides_0, weight = layers_16_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_67_cast_fp16)[name = tensor("op_6380_cast_fp16")]; + tensor var_6386_pad_type_0 = const()[name = tensor("op_6386_pad_type_0"), val = tensor("valid")]; + tensor var_6386_strides_0 = const()[name = tensor("op_6386_strides_0"), val = tensor([1, 1])]; + tensor var_6386_pad_0 = const()[name = tensor("op_6386_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6386_dilations_0 = const()[name = tensor("op_6386_dilations_0"), val = tensor([1, 1])]; + tensor var_6386_groups_0 = const()[name = tensor("op_6386_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(383685120))), name = tensor("layers_16_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(383652992))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6386_cast_fp16 = conv(dilations = var_6386_dilations_0, groups = var_6386_groups_0, pad = var_6386_pad_0, pad_type = var_6386_pad_type_0, strides = var_6386_strides_0, weight = layers_16_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_67_cast_fp16)[name = tensor("op_6386_cast_fp16")]; + tensor value_33_cast_fp16 = add(x = var_6380_cast_fp16, y = var_6386_cast_fp16)[name = tensor("value_33_cast_fp16")]; + tensor var_6389_to_fp16 = const()[name = tensor("op_6389_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(383816256)))]; + tensor query_67_cast_fp16 = add(x = query_65_cast_fp16, y = var_6389_to_fp16)[name = tensor("query_67_cast_fp16")]; + tensor var_6392_to_fp16 = const()[name = tensor("op_6392_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(383818368)))]; + tensor q_with_bias_v_33_cast_fp16 = add(x = query_65_cast_fp16, y = var_6392_to_fp16)[name = tensor("q_with_bias_v_33_cast_fp16")]; + tensor var_6402_pad_type_0 = const()[name = tensor("op_6402_pad_type_0"), val = tensor("valid")]; + tensor var_6402_strides_0 = const()[name = tensor("op_6402_strides_0"), val = tensor([1, 1])]; + tensor var_6402_pad_0 = const()[name = tensor("op_6402_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6402_dilations_0 = const()[name = tensor("op_6402_dilations_0"), val = tensor([1, 1])]; + tensor var_6402_groups_0 = const()[name = tensor("op_6402_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(383820480))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(384606976))), name = tensor("layers_16_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6402_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6402_dilations_0, groups = var_6402_groups_0, pad = var_6402_pad_0, pad_type = var_6402_pad_type_0, strides = var_6402_strides_0, weight = layers_16_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_6402_cast_fp16")]; + tensor var_6408_pad_type_0 = const()[name = tensor("op_6408_pad_type_0"), val = tensor("valid")]; + tensor var_6408_strides_0 = const()[name = tensor("op_6408_strides_0"), val = tensor([1, 1])]; + tensor var_6408_pad_0 = const()[name = tensor("op_6408_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6408_dilations_0 = const()[name = tensor("op_6408_dilations_0"), val = tensor([1, 1])]; + tensor var_6408_groups_0 = const()[name = tensor("op_6408_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(384663680))), name = tensor("layers_16_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(384607168))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6408_cast_fp16 = conv(dilations = var_6408_dilations_0, groups = var_6408_groups_0, pad = var_6408_pad_0, pad_type = var_6408_pad_type_0, strides = var_6408_strides_0, weight = layers_16_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_6408_cast_fp16")]; + tensor p_33_cast_fp16 = add(x = var_6402_cast_fp16, y = var_6408_cast_fp16)[name = tensor("p_33_cast_fp16")]; + tensor var_6412 = const()[name = tensor("op_6412"), val = tensor([1, 8, 128, 188])]; + tensor var_6413_cast_fp16 = reshape(shape = var_6412, x = q_with_bias_v_33_cast_fp16)[name = tensor("op_6413_cast_fp16")]; + tensor var_6414 = const()[name = tensor("op_6414"), val = tensor([1, 8, 128, -1])]; + tensor var_6415_cast_fp16 = reshape(shape = var_6414, x = p_33_cast_fp16)[name = tensor("op_6415_cast_fp16")]; + tensor matrix_bd_129_transpose_x_0 = const()[name = tensor("matrix_bd_129_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_129_transpose_y_0 = const()[name = tensor("matrix_bd_129_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_129_cast_fp16 = matmul(transpose_x = matrix_bd_129_transpose_x_0, transpose_y = matrix_bd_129_transpose_y_0, x = var_6413_cast_fp16, y = var_6415_cast_fp16)[name = tensor("matrix_bd_129_cast_fp16")]; + tensor matrix_bd_131_pad_0 = const()[name = tensor("matrix_bd_131_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_131_mode_0 = const()[name = tensor("matrix_bd_131_mode_0"), val = tensor("constant")]; + tensor const_186_to_fp16 = const()[name = tensor("const_186_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_131_cast_fp16 = pad(constant_val = const_186_to_fp16, mode = matrix_bd_131_mode_0, pad = matrix_bd_131_pad_0, x = matrix_bd_129_cast_fp16)[name = tensor("matrix_bd_131_cast_fp16")]; + tensor var_6424 = const()[name = tensor("op_6424"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_133_cast_fp16 = reshape(shape = var_6424, x = matrix_bd_131_cast_fp16)[name = tensor("matrix_bd_133_cast_fp16")]; + tensor var_6428_begin_0 = const()[name = tensor("op_6428_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_6428_end_0 = const()[name = tensor("op_6428_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_6428_end_mask_0 = const()[name = tensor("op_6428_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_6428_cast_fp16 = slice_by_index(begin = var_6428_begin_0, end = var_6428_end_0, end_mask = var_6428_end_mask_0, x = matrix_bd_133_cast_fp16)[name = tensor("op_6428_cast_fp16")]; + tensor var_6429 = const()[name = tensor("op_6429"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_135_cast_fp16 = reshape(shape = var_6429, x = var_6428_cast_fp16)[name = tensor("matrix_bd_135_cast_fp16")]; + tensor var_6434_begin_0 = const()[name = tensor("op_6434_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6434_end_0 = const()[name = tensor("op_6434_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_6434_end_mask_0 = const()[name = tensor("op_6434_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6434_cast_fp16 = slice_by_index(begin = var_6434_begin_0, end = var_6434_end_0, end_mask = var_6434_end_mask_0, x = matrix_bd_135_cast_fp16)[name = tensor("op_6434_cast_fp16")]; + tensor var_6435_to_fp16 = const()[name = tensor("op_6435_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_33_cast_fp16 = mul(x = var_6434_cast_fp16, y = var_6435_to_fp16)[name = tensor("qk_mask_33_cast_fp16")]; + tensor var_6439 = const()[name = tensor("op_6439"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_33_cast_fp16 = reshape(shape = var_6439, x = query_67_cast_fp16)[name = tensor("mh_q_33_cast_fp16")]; + tensor var_6441_to_fp16 = const()[name = tensor("op_6441_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_6442_cast_fp16 = mul(x = mh_q_33_cast_fp16, y = var_6441_to_fp16)[name = tensor("op_6442_cast_fp16")]; + tensor var_6445 = const()[name = tensor("op_6445"), val = tensor([1, 8, 128, 188])]; + tensor var_6446_cast_fp16 = reshape(shape = var_6445, x = key_33_cast_fp16)[name = tensor("op_6446_cast_fp16")]; + tensor mh_w_65_transpose_x_0 = const()[name = tensor("mh_w_65_transpose_x_0"), val = tensor(true)]; + tensor mh_w_65_transpose_y_0 = const()[name = tensor("mh_w_65_transpose_y_0"), val = tensor(false)]; + tensor mh_w_65_cast_fp16 = matmul(transpose_x = mh_w_65_transpose_x_0, transpose_y = mh_w_65_transpose_y_0, x = var_6442_cast_fp16, y = var_6446_cast_fp16)[name = tensor("mh_w_65_cast_fp16")]; + tensor mh_w_67_cast_fp16 = add(x = mh_w_65_cast_fp16, y = qk_mask_33_cast_fp16)[name = tensor("mh_w_67_cast_fp16")]; + tensor var_6450_cast_fp16 = softmax(axis = var_6237, x = mh_w_67_cast_fp16)[name = tensor("op_6450_cast_fp16")]; + tensor var_6451 = const()[name = tensor("op_6451"), val = tensor([1, 8, 128, 188])]; + tensor var_6452_cast_fp16 = reshape(shape = var_6451, x = value_33_cast_fp16)[name = tensor("op_6452_cast_fp16")]; + tensor attn_33_transpose_x_0 = const()[name = tensor("attn_33_transpose_x_0"), val = tensor(false)]; + tensor attn_33_transpose_y_0 = const()[name = tensor("attn_33_transpose_y_0"), val = tensor(true)]; + tensor attn_33_cast_fp16 = matmul(transpose_x = attn_33_transpose_x_0, transpose_y = attn_33_transpose_y_0, x = var_6452_cast_fp16, y = var_6450_cast_fp16)[name = tensor("attn_33_cast_fp16")]; + tensor var_6455 = const()[name = tensor("op_6455"), val = tensor([1, 1024, 1, 188])]; + tensor input_439_cast_fp16 = reshape(shape = var_6455, x = attn_33_cast_fp16)[name = tensor("input_439_cast_fp16")]; + tensor var_6465_pad_type_0 = const()[name = tensor("op_6465_pad_type_0"), val = tensor("valid")]; + tensor var_6465_strides_0 = const()[name = tensor("op_6465_strides_0"), val = tensor([1, 1])]; + tensor var_6465_pad_0 = const()[name = tensor("op_6465_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6465_dilations_0 = const()[name = tensor("op_6465_dilations_0"), val = tensor([1, 1])]; + tensor var_6465_groups_0 = const()[name = tensor("op_6465_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(384794816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(385581312))), name = tensor("layers_16_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6465_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6465_dilations_0, groups = var_6465_groups_0, pad = var_6465_pad_0, pad_type = var_6465_pad_type_0, strides = var_6465_strides_0, weight = layers_16_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_439_cast_fp16)[name = tensor("op_6465_cast_fp16")]; + tensor var_6471_pad_type_0 = const()[name = tensor("op_6471_pad_type_0"), val = tensor("valid")]; + tensor var_6471_strides_0 = const()[name = tensor("op_6471_strides_0"), val = tensor([1, 1])]; + tensor var_6471_pad_0 = const()[name = tensor("op_6471_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6471_dilations_0 = const()[name = tensor("op_6471_dilations_0"), val = tensor([1, 1])]; + tensor var_6471_groups_0 = const()[name = tensor("op_6471_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(385616000))), name = tensor("layers_16_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(385581504))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6471_cast_fp16 = conv(dilations = var_6471_dilations_0, groups = var_6471_groups_0, pad = var_6471_pad_0, pad_type = var_6471_pad_type_0, strides = var_6471_strides_0, weight = layers_16_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_439_cast_fp16)[name = tensor("op_6471_cast_fp16")]; + tensor obj_69_cast_fp16 = add(x = var_6465_cast_fp16, y = var_6471_cast_fp16)[name = tensor("obj_69_cast_fp16")]; + tensor inputs_165_cast_fp16 = add(x = inputs_163_cast_fp16, y = obj_69_cast_fp16)[name = tensor("inputs_165_cast_fp16")]; + tensor out_165_axes_0 = const()[name = tensor("out_165_axes_0"), val = tensor([1])]; + tensor var_6482_to_fp16 = const()[name = tensor("op_6482_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_165_cast_fp16 = layer_norm(axes = out_165_axes_0, epsilon = var_6482_to_fp16, x = inputs_165_cast_fp16)[name = tensor("out_165_cast_fp16")]; + tensor input_441_gamma_0_to_fp16 = const()[name = tensor("input_441_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(385747136)))]; + tensor input_441_beta_0_to_fp16 = const()[name = tensor("input_441_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(385749248)))]; + tensor input_441_epsilon_0_to_fp16 = const()[name = tensor("input_441_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_441_cast_fp16 = batch_norm(beta = input_441_beta_0_to_fp16, epsilon = input_441_epsilon_0_to_fp16, gamma = input_441_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_165_cast_fp16)[name = tensor("input_441_cast_fp16")]; + tensor var_6503_pad_type_0 = const()[name = tensor("op_6503_pad_type_0"), val = tensor("valid")]; + tensor var_6503_strides_0 = const()[name = tensor("op_6503_strides_0"), val = tensor([1, 1])]; + tensor var_6503_pad_0 = const()[name = tensor("op_6503_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6503_dilations_0 = const()[name = tensor("op_6503_dilations_0"), val = tensor([1, 1])]; + tensor var_6503_groups_0 = const()[name = tensor("op_6503_groups_0"), val = tensor(1)]; + tensor layers_16_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(385751360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387324288))), name = tensor("layers_16_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_6503_cast_fp16 = conv(dilations = var_6503_dilations_0, groups = var_6503_groups_0, pad = var_6503_pad_0, pad_type = var_6503_pad_type_0, strides = var_6503_strides_0, weight = layers_16_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_441_cast_fp16)[name = tensor("op_6503_cast_fp16")]; + tensor var_6509_pad_type_0 = const()[name = tensor("op_6509_pad_type_0"), val = tensor("valid")]; + tensor var_6509_strides_0 = const()[name = tensor("op_6509_strides_0"), val = tensor([1, 1])]; + tensor var_6509_pad_0 = const()[name = tensor("op_6509_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6509_dilations_0 = const()[name = tensor("op_6509_dilations_0"), val = tensor([1, 1])]; + tensor var_6509_groups_0 = const()[name = tensor("op_6509_groups_0"), val = tensor(1)]; + tensor layers_16_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387388096))), name = tensor("layers_16_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387324480))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_6509_cast_fp16 = conv(dilations = var_6509_dilations_0, groups = var_6509_groups_0, pad = var_6509_pad_0, pad_type = var_6509_pad_type_0, strides = var_6509_strides_0, weight = layers_16_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_441_cast_fp16)[name = tensor("op_6509_cast_fp16")]; + tensor input_443_cast_fp16 = add(x = var_6503_cast_fp16, y = var_6509_cast_fp16)[name = tensor("input_443_cast_fp16")]; + tensor input_445_split_num_splits_0 = const()[name = tensor("input_445_split_num_splits_0"), val = tensor(2)]; + tensor input_445_split_axis_0 = const()[name = tensor("input_445_split_axis_0"), val = tensor(1)]; + tensor input_445_split_cast_fp16_0, tensor input_445_split_cast_fp16_1 = split(axis = input_445_split_axis_0, num_splits = input_445_split_num_splits_0, x = input_443_cast_fp16)[name = tensor("input_445_split_cast_fp16")]; + tensor input_445_split_1_sigmoid_cast_fp16 = sigmoid(x = input_445_split_cast_fp16_1)[name = tensor("input_445_split_1_sigmoid_cast_fp16")]; + tensor input_445_cast_fp16 = mul(x = input_445_split_cast_fp16_0, y = input_445_split_1_sigmoid_cast_fp16)[name = tensor("input_445_cast_fp16")]; + tensor input_447_pad_type_0 = const()[name = tensor("input_447_pad_type_0"), val = tensor("custom")]; + tensor input_447_pad_0 = const()[name = tensor("input_447_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_447_groups_0 = const()[name = tensor("input_447_groups_0"), val = tensor(1024)]; + tensor input_447_strides_0 = const()[name = tensor("input_447_strides_0"), val = tensor([1, 1])]; + tensor input_447_dilations_0 = const()[name = tensor("input_447_dilations_0"), val = tensor([1, 1])]; + tensor const_300_to_fp16 = const()[name = tensor("const_300_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387650304)))]; + tensor const_301_to_fp16 = const()[name = tensor("const_301_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387668800)))]; + tensor input_449_cast_fp16 = conv(bias = const_301_to_fp16, dilations = input_447_dilations_0, groups = input_447_groups_0, pad = input_447_pad_0, pad_type = input_447_pad_type_0, strides = input_447_strides_0, weight = const_300_to_fp16, x = input_445_cast_fp16)[name = tensor("input_449_cast_fp16")]; + tensor input_451_cast_fp16 = silu(x = input_449_cast_fp16)[name = tensor("input_451_cast_fp16")]; + tensor var_6531_pad_type_0 = const()[name = tensor("op_6531_pad_type_0"), val = tensor("valid")]; + tensor var_6531_strides_0 = const()[name = tensor("op_6531_strides_0"), val = tensor([1, 1])]; + tensor var_6531_pad_0 = const()[name = tensor("op_6531_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6531_dilations_0 = const()[name = tensor("op_6531_dilations_0"), val = tensor([1, 1])]; + tensor var_6531_groups_0 = const()[name = tensor("op_6531_groups_0"), val = tensor(1)]; + tensor layers_16_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387670912))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(388457408))), name = tensor("layers_16_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6531_cast_fp16 = conv(dilations = var_6531_dilations_0, groups = var_6531_groups_0, pad = var_6531_pad_0, pad_type = var_6531_pad_type_0, strides = var_6531_strides_0, weight = layers_16_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_451_cast_fp16)[name = tensor("op_6531_cast_fp16")]; + tensor var_6537_pad_type_0 = const()[name = tensor("op_6537_pad_type_0"), val = tensor("valid")]; + tensor var_6537_strides_0 = const()[name = tensor("op_6537_strides_0"), val = tensor([1, 1])]; + tensor var_6537_pad_0 = const()[name = tensor("op_6537_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6537_dilations_0 = const()[name = tensor("op_6537_dilations_0"), val = tensor([1, 1])]; + tensor var_6537_groups_0 = const()[name = tensor("op_6537_groups_0"), val = tensor(1)]; + tensor layers_16_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(388488256))), name = tensor("layers_16_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(388457600))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6537_cast_fp16 = conv(dilations = var_6537_dilations_0, groups = var_6537_groups_0, pad = var_6537_pad_0, pad_type = var_6537_pad_type_0, strides = var_6537_strides_0, weight = layers_16_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_451_cast_fp16)[name = tensor("op_6537_cast_fp16")]; + tensor x_101_cast_fp16 = add(x = var_6531_cast_fp16, y = var_6537_cast_fp16)[name = tensor("x_101_cast_fp16")]; + tensor inputs_167_cast_fp16 = add(x = inputs_165_cast_fp16, y = x_101_cast_fp16)[name = tensor("inputs_167_cast_fp16")]; + tensor out_167_axes_0 = const()[name = tensor("out_167_axes_0"), val = tensor([1])]; + tensor var_6548_to_fp16 = const()[name = tensor("op_6548_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_167_cast_fp16 = layer_norm(axes = out_167_axes_0, epsilon = var_6548_to_fp16, x = inputs_167_cast_fp16)[name = tensor("out_167_cast_fp16")]; + tensor input_453_gamma_0_to_fp16 = const()[name = tensor("input_453_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(388619392)))]; + tensor input_453_beta_0_to_fp16 = const()[name = tensor("input_453_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(388621504)))]; + tensor input_453_epsilon_0_to_fp16 = const()[name = tensor("input_453_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_453_cast_fp16 = batch_norm(beta = input_453_beta_0_to_fp16, epsilon = input_453_epsilon_0_to_fp16, gamma = input_453_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_167_cast_fp16)[name = tensor("input_453_cast_fp16")]; + tensor var_6568_pad_type_0 = const()[name = tensor("op_6568_pad_type_0"), val = tensor("valid")]; + tensor var_6568_strides_0 = const()[name = tensor("op_6568_strides_0"), val = tensor([1, 1])]; + tensor var_6568_pad_0 = const()[name = tensor("op_6568_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6568_dilations_0 = const()[name = tensor("op_6568_dilations_0"), val = tensor([1, 1])]; + tensor var_6568_groups_0 = const()[name = tensor("op_6568_groups_0"), val = tensor(1)]; + tensor layers_16_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(388623616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(391769408))), name = tensor("layers_16_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_6568_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_6568_dilations_0, groups = var_6568_groups_0, pad = var_6568_pad_0, pad_type = var_6568_pad_type_0, strides = var_6568_strides_0, weight = layers_16_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_453_cast_fp16)[name = tensor("op_6568_cast_fp16")]; + tensor var_6574_pad_type_0 = const()[name = tensor("op_6574_pad_type_0"), val = tensor("valid")]; + tensor var_6574_strides_0 = const()[name = tensor("op_6574_strides_0"), val = tensor([1, 1])]; + tensor var_6574_pad_0 = const()[name = tensor("op_6574_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6574_dilations_0 = const()[name = tensor("op_6574_dilations_0"), val = tensor([1, 1])]; + tensor var_6574_groups_0 = const()[name = tensor("op_6574_groups_0"), val = tensor(1)]; + tensor layers_16_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(391905728))), name = tensor("layers_16_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(391769600))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_6574_cast_fp16 = conv(dilations = var_6574_dilations_0, groups = var_6574_groups_0, pad = var_6574_pad_0, pad_type = var_6574_pad_type_0, strides = var_6574_strides_0, weight = layers_16_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_453_cast_fp16)[name = tensor("op_6574_cast_fp16")]; + tensor input_455_cast_fp16 = add(x = var_6568_cast_fp16, y = var_6574_cast_fp16)[name = tensor("input_455_cast_fp16")]; + tensor input_457_cast_fp16 = silu(x = input_455_cast_fp16)[name = tensor("input_457_cast_fp16")]; + tensor var_6585_pad_type_0 = const()[name = tensor("op_6585_pad_type_0"), val = tensor("valid")]; + tensor var_6585_strides_0 = const()[name = tensor("op_6585_strides_0"), val = tensor([1, 1])]; + tensor var_6585_pad_0 = const()[name = tensor("op_6585_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6585_dilations_0 = const()[name = tensor("op_6585_dilations_0"), val = tensor([1, 1])]; + tensor var_6585_groups_0 = const()[name = tensor("op_6585_groups_0"), val = tensor(1)]; + tensor layers_16_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(392430080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(395575872))), name = tensor("layers_16_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_6585_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6585_dilations_0, groups = var_6585_groups_0, pad = var_6585_pad_0, pad_type = var_6585_pad_type_0, strides = var_6585_strides_0, weight = layers_16_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_457_cast_fp16)[name = tensor("op_6585_cast_fp16")]; + tensor var_6591_pad_type_0 = const()[name = tensor("op_6591_pad_type_0"), val = tensor("valid")]; + tensor var_6591_strides_0 = const()[name = tensor("op_6591_strides_0"), val = tensor([1, 1])]; + tensor var_6591_pad_0 = const()[name = tensor("op_6591_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6591_dilations_0 = const()[name = tensor("op_6591_dilations_0"), val = tensor([1, 1])]; + tensor var_6591_groups_0 = const()[name = tensor("op_6591_groups_0"), val = tensor(1)]; + tensor layers_16_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(395716032))), name = tensor("layers_16_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(395576064))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_6591_cast_fp16 = conv(dilations = var_6591_dilations_0, groups = var_6591_groups_0, pad = var_6591_pad_0, pad_type = var_6591_pad_type_0, strides = var_6591_strides_0, weight = layers_16_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_457_cast_fp16)[name = tensor("op_6591_cast_fp16")]; + tensor x_103_cast_fp16 = add(x = var_6585_cast_fp16, y = var_6591_cast_fp16)[name = tensor("x_103_cast_fp16")]; + tensor var_6593_to_fp16 = const()[name = tensor("op_6593_to_fp16"), val = tensor(0x1p-1)]; + tensor var_6594_cast_fp16 = mul(x = x_103_cast_fp16, y = var_6593_to_fp16)[name = tensor("op_6594_cast_fp16")]; + tensor inputs_169_cast_fp16 = add(x = inputs_167_cast_fp16, y = var_6594_cast_fp16)[name = tensor("inputs_169_cast_fp16")]; + tensor out_169_axes_0 = const()[name = tensor("out_169_axes_0"), val = tensor([1])]; + tensor var_6604_to_fp16 = const()[name = tensor("op_6604_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_169_cast_fp16 = layer_norm(axes = out_169_axes_0, epsilon = var_6604_to_fp16, x = inputs_169_cast_fp16)[name = tensor("out_169_cast_fp16")]; + tensor inputs_171_gamma_0_to_fp16 = const()[name = tensor("inputs_171_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396240384)))]; + tensor inputs_171_beta_0_to_fp16 = const()[name = tensor("inputs_171_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396242496)))]; + tensor inputs_171_epsilon_0_to_fp16 = const()[name = tensor("inputs_171_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_171_cast_fp16 = batch_norm(beta = inputs_171_beta_0_to_fp16, epsilon = inputs_171_epsilon_0_to_fp16, gamma = inputs_171_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_169_cast_fp16)[name = tensor("inputs_171_cast_fp16")]; + tensor var_6618 = const()[name = tensor("op_6618"), val = tensor(3)]; + tensor out_171_axes_0 = const()[name = tensor("out_171_axes_0"), val = tensor([1])]; + tensor var_6649_to_fp16 = const()[name = tensor("op_6649_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_171_cast_fp16 = layer_norm(axes = out_171_axes_0, epsilon = var_6649_to_fp16, x = inputs_171_cast_fp16)[name = tensor("out_171_cast_fp16")]; + tensor input_459_gamma_0_to_fp16 = const()[name = tensor("input_459_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396244608)))]; + tensor input_459_beta_0_to_fp16 = const()[name = tensor("input_459_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396246720)))]; + tensor input_459_epsilon_0_to_fp16 = const()[name = tensor("input_459_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_459_cast_fp16 = batch_norm(beta = input_459_beta_0_to_fp16, epsilon = input_459_epsilon_0_to_fp16, gamma = input_459_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_171_cast_fp16)[name = tensor("input_459_cast_fp16")]; + tensor var_6669_pad_type_0 = const()[name = tensor("op_6669_pad_type_0"), val = tensor("valid")]; + tensor var_6669_strides_0 = const()[name = tensor("op_6669_strides_0"), val = tensor([1, 1])]; + tensor var_6669_pad_0 = const()[name = tensor("op_6669_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6669_dilations_0 = const()[name = tensor("op_6669_dilations_0"), val = tensor([1, 1])]; + tensor var_6669_groups_0 = const()[name = tensor("op_6669_groups_0"), val = tensor(1)]; + tensor layers_17_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396248832))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(399394624))), name = tensor("layers_17_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_6669_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_6669_dilations_0, groups = var_6669_groups_0, pad = var_6669_pad_0, pad_type = var_6669_pad_type_0, strides = var_6669_strides_0, weight = layers_17_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_459_cast_fp16)[name = tensor("op_6669_cast_fp16")]; + tensor var_6675_pad_type_0 = const()[name = tensor("op_6675_pad_type_0"), val = tensor("valid")]; + tensor var_6675_strides_0 = const()[name = tensor("op_6675_strides_0"), val = tensor([1, 1])]; + tensor var_6675_pad_0 = const()[name = tensor("op_6675_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6675_dilations_0 = const()[name = tensor("op_6675_dilations_0"), val = tensor([1, 1])]; + tensor var_6675_groups_0 = const()[name = tensor("op_6675_groups_0"), val = tensor(1)]; + tensor layers_17_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(399539200))), name = tensor("layers_17_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(399394816))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_6675_cast_fp16 = conv(dilations = var_6675_dilations_0, groups = var_6675_groups_0, pad = var_6675_pad_0, pad_type = var_6675_pad_type_0, strides = var_6675_strides_0, weight = layers_17_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_459_cast_fp16)[name = tensor("op_6675_cast_fp16")]; + tensor input_461_cast_fp16 = add(x = var_6669_cast_fp16, y = var_6675_cast_fp16)[name = tensor("input_461_cast_fp16")]; + tensor input_463_cast_fp16 = silu(x = input_461_cast_fp16)[name = tensor("input_463_cast_fp16")]; + tensor var_6686_pad_type_0 = const()[name = tensor("op_6686_pad_type_0"), val = tensor("valid")]; + tensor var_6686_strides_0 = const()[name = tensor("op_6686_strides_0"), val = tensor([1, 1])]; + tensor var_6686_pad_0 = const()[name = tensor("op_6686_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6686_dilations_0 = const()[name = tensor("op_6686_dilations_0"), val = tensor([1, 1])]; + tensor var_6686_groups_0 = const()[name = tensor("op_6686_groups_0"), val = tensor(1)]; + tensor layers_17_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(400063552))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(403209344))), name = tensor("layers_17_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_6686_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6686_dilations_0, groups = var_6686_groups_0, pad = var_6686_pad_0, pad_type = var_6686_pad_type_0, strides = var_6686_strides_0, weight = layers_17_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_463_cast_fp16)[name = tensor("op_6686_cast_fp16")]; + tensor var_6692_pad_type_0 = const()[name = tensor("op_6692_pad_type_0"), val = tensor("valid")]; + tensor var_6692_strides_0 = const()[name = tensor("op_6692_strides_0"), val = tensor([1, 1])]; + tensor var_6692_pad_0 = const()[name = tensor("op_6692_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6692_dilations_0 = const()[name = tensor("op_6692_dilations_0"), val = tensor([1, 1])]; + tensor var_6692_groups_0 = const()[name = tensor("op_6692_groups_0"), val = tensor(1)]; + tensor layers_17_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(403354496))), name = tensor("layers_17_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(403209536))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_6692_cast_fp16 = conv(dilations = var_6692_dilations_0, groups = var_6692_groups_0, pad = var_6692_pad_0, pad_type = var_6692_pad_type_0, strides = var_6692_strides_0, weight = layers_17_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_463_cast_fp16)[name = tensor("op_6692_cast_fp16")]; + tensor x_105_cast_fp16 = add(x = var_6686_cast_fp16, y = var_6692_cast_fp16)[name = tensor("x_105_cast_fp16")]; + tensor var_6694_to_fp16 = const()[name = tensor("op_6694_to_fp16"), val = tensor(0x1p-1)]; + tensor var_6695_cast_fp16 = mul(x = x_105_cast_fp16, y = var_6694_to_fp16)[name = tensor("op_6695_cast_fp16")]; + tensor inputs_173_cast_fp16 = add(x = inputs_171_cast_fp16, y = var_6695_cast_fp16)[name = tensor("inputs_173_cast_fp16")]; + tensor out_173_axes_0 = const()[name = tensor("out_173_axes_0"), val = tensor([1])]; + tensor var_6705_to_fp16 = const()[name = tensor("op_6705_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_173_cast_fp16 = layer_norm(axes = out_173_axes_0, epsilon = var_6705_to_fp16, x = inputs_173_cast_fp16)[name = tensor("out_173_cast_fp16")]; + tensor obj_71_gamma_0_to_fp16 = const()[name = tensor("obj_71_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(403878848)))]; + tensor obj_71_beta_0_to_fp16 = const()[name = tensor("obj_71_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(403880960)))]; + tensor obj_71_epsilon_0_to_fp16 = const()[name = tensor("obj_71_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_71_cast_fp16 = batch_norm(beta = obj_71_beta_0_to_fp16, epsilon = obj_71_epsilon_0_to_fp16, gamma = obj_71_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_173_cast_fp16)[name = tensor("obj_71_cast_fp16")]; + tensor var_6730_pad_type_0 = const()[name = tensor("op_6730_pad_type_0"), val = tensor("valid")]; + tensor var_6730_strides_0 = const()[name = tensor("op_6730_strides_0"), val = tensor([1, 1])]; + tensor var_6730_pad_0 = const()[name = tensor("op_6730_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6730_dilations_0 = const()[name = tensor("op_6730_dilations_0"), val = tensor([1, 1])]; + tensor var_6730_groups_0 = const()[name = tensor("op_6730_groups_0"), val = tensor(1)]; + tensor layers_17_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(403883072))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(404669568))), name = tensor("layers_17_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6730_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6730_dilations_0, groups = var_6730_groups_0, pad = var_6730_pad_0, pad_type = var_6730_pad_type_0, strides = var_6730_strides_0, weight = layers_17_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_71_cast_fp16)[name = tensor("op_6730_cast_fp16")]; + tensor var_6736_pad_type_0 = const()[name = tensor("op_6736_pad_type_0"), val = tensor("valid")]; + tensor var_6736_strides_0 = const()[name = tensor("op_6736_strides_0"), val = tensor([1, 1])]; + tensor var_6736_pad_0 = const()[name = tensor("op_6736_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6736_dilations_0 = const()[name = tensor("op_6736_dilations_0"), val = tensor([1, 1])]; + tensor var_6736_groups_0 = const()[name = tensor("op_6736_groups_0"), val = tensor(1)]; + tensor layers_17_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(404700928))), name = tensor("layers_17_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(404669760))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6736_cast_fp16 = conv(dilations = var_6736_dilations_0, groups = var_6736_groups_0, pad = var_6736_pad_0, pad_type = var_6736_pad_type_0, strides = var_6736_strides_0, weight = layers_17_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_71_cast_fp16)[name = tensor("op_6736_cast_fp16")]; + tensor query_69_cast_fp16 = add(x = var_6730_cast_fp16, y = var_6736_cast_fp16)[name = tensor("query_69_cast_fp16")]; + tensor var_6745_pad_type_0 = const()[name = tensor("op_6745_pad_type_0"), val = tensor("valid")]; + tensor var_6745_strides_0 = const()[name = tensor("op_6745_strides_0"), val = tensor([1, 1])]; + tensor var_6745_pad_0 = const()[name = tensor("op_6745_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6745_dilations_0 = const()[name = tensor("op_6745_dilations_0"), val = tensor([1, 1])]; + tensor var_6745_groups_0 = const()[name = tensor("op_6745_groups_0"), val = tensor(1)]; + tensor layers_17_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(404832064))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(405618560))), name = tensor("layers_17_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6745_cast_fp16 = conv(dilations = var_6745_dilations_0, groups = var_6745_groups_0, pad = var_6745_pad_0, pad_type = var_6745_pad_type_0, strides = var_6745_strides_0, weight = layers_17_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_71_cast_fp16)[name = tensor("op_6745_cast_fp16")]; + tensor var_6751_pad_type_0 = const()[name = tensor("op_6751_pad_type_0"), val = tensor("valid")]; + tensor var_6751_strides_0 = const()[name = tensor("op_6751_strides_0"), val = tensor([1, 1])]; + tensor var_6751_pad_0 = const()[name = tensor("op_6751_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6751_dilations_0 = const()[name = tensor("op_6751_dilations_0"), val = tensor([1, 1])]; + tensor var_6751_groups_0 = const()[name = tensor("op_6751_groups_0"), val = tensor(1)]; + tensor layers_17_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(405654464))), name = tensor("layers_17_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(405618752))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6751_cast_fp16 = conv(dilations = var_6751_dilations_0, groups = var_6751_groups_0, pad = var_6751_pad_0, pad_type = var_6751_pad_type_0, strides = var_6751_strides_0, weight = layers_17_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_71_cast_fp16)[name = tensor("op_6751_cast_fp16")]; + tensor key_35_cast_fp16 = add(x = var_6745_cast_fp16, y = var_6751_cast_fp16)[name = tensor("key_35_cast_fp16")]; + tensor var_6761_pad_type_0 = const()[name = tensor("op_6761_pad_type_0"), val = tensor("valid")]; + tensor var_6761_strides_0 = const()[name = tensor("op_6761_strides_0"), val = tensor([1, 1])]; + tensor var_6761_pad_0 = const()[name = tensor("op_6761_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6761_dilations_0 = const()[name = tensor("op_6761_dilations_0"), val = tensor([1, 1])]; + tensor var_6761_groups_0 = const()[name = tensor("op_6761_groups_0"), val = tensor(1)]; + tensor layers_17_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(405785600))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(406572096))), name = tensor("layers_17_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6761_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6761_dilations_0, groups = var_6761_groups_0, pad = var_6761_pad_0, pad_type = var_6761_pad_type_0, strides = var_6761_strides_0, weight = layers_17_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_71_cast_fp16)[name = tensor("op_6761_cast_fp16")]; + tensor var_6767_pad_type_0 = const()[name = tensor("op_6767_pad_type_0"), val = tensor("valid")]; + tensor var_6767_strides_0 = const()[name = tensor("op_6767_strides_0"), val = tensor([1, 1])]; + tensor var_6767_pad_0 = const()[name = tensor("op_6767_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6767_dilations_0 = const()[name = tensor("op_6767_dilations_0"), val = tensor([1, 1])]; + tensor var_6767_groups_0 = const()[name = tensor("op_6767_groups_0"), val = tensor(1)]; + tensor layers_17_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(406603968))), name = tensor("layers_17_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(406572288))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6767_cast_fp16 = conv(dilations = var_6767_dilations_0, groups = var_6767_groups_0, pad = var_6767_pad_0, pad_type = var_6767_pad_type_0, strides = var_6767_strides_0, weight = layers_17_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_71_cast_fp16)[name = tensor("op_6767_cast_fp16")]; + tensor value_35_cast_fp16 = add(x = var_6761_cast_fp16, y = var_6767_cast_fp16)[name = tensor("value_35_cast_fp16")]; + tensor var_6770_to_fp16 = const()[name = tensor("op_6770_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(406735104)))]; + tensor query_71_cast_fp16 = add(x = query_69_cast_fp16, y = var_6770_to_fp16)[name = tensor("query_71_cast_fp16")]; + tensor var_6773_to_fp16 = const()[name = tensor("op_6773_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(406737216)))]; + tensor q_with_bias_v_35_cast_fp16 = add(x = query_69_cast_fp16, y = var_6773_to_fp16)[name = tensor("q_with_bias_v_35_cast_fp16")]; + tensor var_6783_pad_type_0 = const()[name = tensor("op_6783_pad_type_0"), val = tensor("valid")]; + tensor var_6783_strides_0 = const()[name = tensor("op_6783_strides_0"), val = tensor([1, 1])]; + tensor var_6783_pad_0 = const()[name = tensor("op_6783_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6783_dilations_0 = const()[name = tensor("op_6783_dilations_0"), val = tensor([1, 1])]; + tensor var_6783_groups_0 = const()[name = tensor("op_6783_groups_0"), val = tensor(1)]; + tensor layers_17_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(406739328))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407525824))), name = tensor("layers_17_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6783_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6783_dilations_0, groups = var_6783_groups_0, pad = var_6783_pad_0, pad_type = var_6783_pad_type_0, strides = var_6783_strides_0, weight = layers_17_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_6783_cast_fp16")]; + tensor var_6789_pad_type_0 = const()[name = tensor("op_6789_pad_type_0"), val = tensor("valid")]; + tensor var_6789_strides_0 = const()[name = tensor("op_6789_strides_0"), val = tensor([1, 1])]; + tensor var_6789_pad_0 = const()[name = tensor("op_6789_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6789_dilations_0 = const()[name = tensor("op_6789_dilations_0"), val = tensor([1, 1])]; + tensor var_6789_groups_0 = const()[name = tensor("op_6789_groups_0"), val = tensor(1)]; + tensor layers_17_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407581056))), name = tensor("layers_17_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407526016))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6789_cast_fp16 = conv(dilations = var_6789_dilations_0, groups = var_6789_groups_0, pad = var_6789_pad_0, pad_type = var_6789_pad_type_0, strides = var_6789_strides_0, weight = layers_17_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_6789_cast_fp16")]; + tensor p_35_cast_fp16 = add(x = var_6783_cast_fp16, y = var_6789_cast_fp16)[name = tensor("p_35_cast_fp16")]; + tensor var_6793 = const()[name = tensor("op_6793"), val = tensor([1, 8, 128, 188])]; + tensor var_6794_cast_fp16 = reshape(shape = var_6793, x = q_with_bias_v_35_cast_fp16)[name = tensor("op_6794_cast_fp16")]; + tensor var_6795 = const()[name = tensor("op_6795"), val = tensor([1, 8, 128, -1])]; + tensor var_6796_cast_fp16 = reshape(shape = var_6795, x = p_35_cast_fp16)[name = tensor("op_6796_cast_fp16")]; + tensor matrix_bd_137_transpose_x_0 = const()[name = tensor("matrix_bd_137_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_137_transpose_y_0 = const()[name = tensor("matrix_bd_137_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_137_cast_fp16 = matmul(transpose_x = matrix_bd_137_transpose_x_0, transpose_y = matrix_bd_137_transpose_y_0, x = var_6794_cast_fp16, y = var_6796_cast_fp16)[name = tensor("matrix_bd_137_cast_fp16")]; + tensor matrix_bd_139_pad_0 = const()[name = tensor("matrix_bd_139_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_139_mode_0 = const()[name = tensor("matrix_bd_139_mode_0"), val = tensor("constant")]; + tensor const_197_to_fp16 = const()[name = tensor("const_197_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_139_cast_fp16 = pad(constant_val = const_197_to_fp16, mode = matrix_bd_139_mode_0, pad = matrix_bd_139_pad_0, x = matrix_bd_137_cast_fp16)[name = tensor("matrix_bd_139_cast_fp16")]; + tensor var_6805 = const()[name = tensor("op_6805"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_141_cast_fp16 = reshape(shape = var_6805, x = matrix_bd_139_cast_fp16)[name = tensor("matrix_bd_141_cast_fp16")]; + tensor var_6809_begin_0 = const()[name = tensor("op_6809_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_6809_end_0 = const()[name = tensor("op_6809_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_6809_end_mask_0 = const()[name = tensor("op_6809_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_6809_cast_fp16 = slice_by_index(begin = var_6809_begin_0, end = var_6809_end_0, end_mask = var_6809_end_mask_0, x = matrix_bd_141_cast_fp16)[name = tensor("op_6809_cast_fp16")]; + tensor var_6810 = const()[name = tensor("op_6810"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_143_cast_fp16 = reshape(shape = var_6810, x = var_6809_cast_fp16)[name = tensor("matrix_bd_143_cast_fp16")]; + tensor var_6815_begin_0 = const()[name = tensor("op_6815_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6815_end_0 = const()[name = tensor("op_6815_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_6815_end_mask_0 = const()[name = tensor("op_6815_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6815_cast_fp16 = slice_by_index(begin = var_6815_begin_0, end = var_6815_end_0, end_mask = var_6815_end_mask_0, x = matrix_bd_143_cast_fp16)[name = tensor("op_6815_cast_fp16")]; + tensor var_6816_to_fp16 = const()[name = tensor("op_6816_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_35_cast_fp16 = mul(x = var_6815_cast_fp16, y = var_6816_to_fp16)[name = tensor("qk_mask_35_cast_fp16")]; + tensor var_6820 = const()[name = tensor("op_6820"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_35_cast_fp16 = reshape(shape = var_6820, x = query_71_cast_fp16)[name = tensor("mh_q_35_cast_fp16")]; + tensor var_6822_to_fp16 = const()[name = tensor("op_6822_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_6823_cast_fp16 = mul(x = mh_q_35_cast_fp16, y = var_6822_to_fp16)[name = tensor("op_6823_cast_fp16")]; + tensor var_6826 = const()[name = tensor("op_6826"), val = tensor([1, 8, 128, 188])]; + tensor var_6827_cast_fp16 = reshape(shape = var_6826, x = key_35_cast_fp16)[name = tensor("op_6827_cast_fp16")]; + tensor mh_w_69_transpose_x_0 = const()[name = tensor("mh_w_69_transpose_x_0"), val = tensor(true)]; + tensor mh_w_69_transpose_y_0 = const()[name = tensor("mh_w_69_transpose_y_0"), val = tensor(false)]; + tensor mh_w_69_cast_fp16 = matmul(transpose_x = mh_w_69_transpose_x_0, transpose_y = mh_w_69_transpose_y_0, x = var_6823_cast_fp16, y = var_6827_cast_fp16)[name = tensor("mh_w_69_cast_fp16")]; + tensor mh_w_71_cast_fp16 = add(x = mh_w_69_cast_fp16, y = qk_mask_35_cast_fp16)[name = tensor("mh_w_71_cast_fp16")]; + tensor var_6831_cast_fp16 = softmax(axis = var_6618, x = mh_w_71_cast_fp16)[name = tensor("op_6831_cast_fp16")]; + tensor var_6832 = const()[name = tensor("op_6832"), val = tensor([1, 8, 128, 188])]; + tensor var_6833_cast_fp16 = reshape(shape = var_6832, x = value_35_cast_fp16)[name = tensor("op_6833_cast_fp16")]; + tensor attn_35_transpose_x_0 = const()[name = tensor("attn_35_transpose_x_0"), val = tensor(false)]; + tensor attn_35_transpose_y_0 = const()[name = tensor("attn_35_transpose_y_0"), val = tensor(true)]; + tensor attn_35_cast_fp16 = matmul(transpose_x = attn_35_transpose_x_0, transpose_y = attn_35_transpose_y_0, x = var_6833_cast_fp16, y = var_6831_cast_fp16)[name = tensor("attn_35_cast_fp16")]; + tensor var_6836 = const()[name = tensor("op_6836"), val = tensor([1, 1024, 1, 188])]; + tensor input_465_cast_fp16 = reshape(shape = var_6836, x = attn_35_cast_fp16)[name = tensor("input_465_cast_fp16")]; + tensor var_6846_pad_type_0 = const()[name = tensor("op_6846_pad_type_0"), val = tensor("valid")]; + tensor var_6846_strides_0 = const()[name = tensor("op_6846_strides_0"), val = tensor([1, 1])]; + tensor var_6846_pad_0 = const()[name = tensor("op_6846_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6846_dilations_0 = const()[name = tensor("op_6846_dilations_0"), val = tensor([1, 1])]; + tensor var_6846_groups_0 = const()[name = tensor("op_6846_groups_0"), val = tensor(1)]; + tensor layers_17_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407712192))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408498688))), name = tensor("layers_17_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6846_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6846_dilations_0, groups = var_6846_groups_0, pad = var_6846_pad_0, pad_type = var_6846_pad_type_0, strides = var_6846_strides_0, weight = layers_17_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_465_cast_fp16)[name = tensor("op_6846_cast_fp16")]; + tensor var_6852_pad_type_0 = const()[name = tensor("op_6852_pad_type_0"), val = tensor("valid")]; + tensor var_6852_strides_0 = const()[name = tensor("op_6852_strides_0"), val = tensor([1, 1])]; + tensor var_6852_pad_0 = const()[name = tensor("op_6852_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6852_dilations_0 = const()[name = tensor("op_6852_dilations_0"), val = tensor([1, 1])]; + tensor var_6852_groups_0 = const()[name = tensor("op_6852_groups_0"), val = tensor(1)]; + tensor layers_17_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408530432))), name = tensor("layers_17_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408498880))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6852_cast_fp16 = conv(dilations = var_6852_dilations_0, groups = var_6852_groups_0, pad = var_6852_pad_0, pad_type = var_6852_pad_type_0, strides = var_6852_strides_0, weight = layers_17_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_465_cast_fp16)[name = tensor("op_6852_cast_fp16")]; + tensor obj_73_cast_fp16 = add(x = var_6846_cast_fp16, y = var_6852_cast_fp16)[name = tensor("obj_73_cast_fp16")]; + tensor inputs_175_cast_fp16 = add(x = inputs_173_cast_fp16, y = obj_73_cast_fp16)[name = tensor("inputs_175_cast_fp16")]; + tensor out_175_axes_0 = const()[name = tensor("out_175_axes_0"), val = tensor([1])]; + tensor var_6863_to_fp16 = const()[name = tensor("op_6863_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_175_cast_fp16 = layer_norm(axes = out_175_axes_0, epsilon = var_6863_to_fp16, x = inputs_175_cast_fp16)[name = tensor("out_175_cast_fp16")]; + tensor input_467_gamma_0_to_fp16 = const()[name = tensor("input_467_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408661568)))]; + tensor input_467_beta_0_to_fp16 = const()[name = tensor("input_467_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408663680)))]; + tensor input_467_epsilon_0_to_fp16 = const()[name = tensor("input_467_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_467_cast_fp16 = batch_norm(beta = input_467_beta_0_to_fp16, epsilon = input_467_epsilon_0_to_fp16, gamma = input_467_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_175_cast_fp16)[name = tensor("input_467_cast_fp16")]; + tensor var_6884_pad_type_0 = const()[name = tensor("op_6884_pad_type_0"), val = tensor("valid")]; + tensor var_6884_strides_0 = const()[name = tensor("op_6884_strides_0"), val = tensor([1, 1])]; + tensor var_6884_pad_0 = const()[name = tensor("op_6884_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6884_dilations_0 = const()[name = tensor("op_6884_dilations_0"), val = tensor([1, 1])]; + tensor var_6884_groups_0 = const()[name = tensor("op_6884_groups_0"), val = tensor(1)]; + tensor layers_17_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408665792))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(410238720))), name = tensor("layers_17_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_6884_cast_fp16 = conv(dilations = var_6884_dilations_0, groups = var_6884_groups_0, pad = var_6884_pad_0, pad_type = var_6884_pad_type_0, strides = var_6884_strides_0, weight = layers_17_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_467_cast_fp16)[name = tensor("op_6884_cast_fp16")]; + tensor var_6890_pad_type_0 = const()[name = tensor("op_6890_pad_type_0"), val = tensor("valid")]; + tensor var_6890_strides_0 = const()[name = tensor("op_6890_strides_0"), val = tensor([1, 1])]; + tensor var_6890_pad_0 = const()[name = tensor("op_6890_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6890_dilations_0 = const()[name = tensor("op_6890_dilations_0"), val = tensor([1, 1])]; + tensor var_6890_groups_0 = const()[name = tensor("op_6890_groups_0"), val = tensor(1)]; + tensor layers_17_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(410303680))), name = tensor("layers_17_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(410238912))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_6890_cast_fp16 = conv(dilations = var_6890_dilations_0, groups = var_6890_groups_0, pad = var_6890_pad_0, pad_type = var_6890_pad_type_0, strides = var_6890_strides_0, weight = layers_17_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_467_cast_fp16)[name = tensor("op_6890_cast_fp16")]; + tensor input_469_cast_fp16 = add(x = var_6884_cast_fp16, y = var_6890_cast_fp16)[name = tensor("input_469_cast_fp16")]; + tensor input_471_split_num_splits_0 = const()[name = tensor("input_471_split_num_splits_0"), val = tensor(2)]; + tensor input_471_split_axis_0 = const()[name = tensor("input_471_split_axis_0"), val = tensor(1)]; + tensor input_471_split_cast_fp16_0, tensor input_471_split_cast_fp16_1 = split(axis = input_471_split_axis_0, num_splits = input_471_split_num_splits_0, x = input_469_cast_fp16)[name = tensor("input_471_split_cast_fp16")]; + tensor input_471_split_1_sigmoid_cast_fp16 = sigmoid(x = input_471_split_cast_fp16_1)[name = tensor("input_471_split_1_sigmoid_cast_fp16")]; + tensor input_471_cast_fp16 = mul(x = input_471_split_cast_fp16_0, y = input_471_split_1_sigmoid_cast_fp16)[name = tensor("input_471_cast_fp16")]; + tensor input_473_pad_type_0 = const()[name = tensor("input_473_pad_type_0"), val = tensor("custom")]; + tensor input_473_pad_0 = const()[name = tensor("input_473_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_473_groups_0 = const()[name = tensor("input_473_groups_0"), val = tensor(1024)]; + tensor input_473_strides_0 = const()[name = tensor("input_473_strides_0"), val = tensor([1, 1])]; + tensor input_473_dilations_0 = const()[name = tensor("input_473_dilations_0"), val = tensor([1, 1])]; + tensor const_302_to_fp16 = const()[name = tensor("const_302_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(410565888)))]; + tensor const_303_to_fp16 = const()[name = tensor("const_303_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(410584384)))]; + tensor input_475_cast_fp16 = conv(bias = const_303_to_fp16, dilations = input_473_dilations_0, groups = input_473_groups_0, pad = input_473_pad_0, pad_type = input_473_pad_type_0, strides = input_473_strides_0, weight = const_302_to_fp16, x = input_471_cast_fp16)[name = tensor("input_475_cast_fp16")]; + tensor input_477_cast_fp16 = silu(x = input_475_cast_fp16)[name = tensor("input_477_cast_fp16")]; + tensor var_6912_pad_type_0 = const()[name = tensor("op_6912_pad_type_0"), val = tensor("valid")]; + tensor var_6912_strides_0 = const()[name = tensor("op_6912_strides_0"), val = tensor([1, 1])]; + tensor var_6912_pad_0 = const()[name = tensor("op_6912_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6912_dilations_0 = const()[name = tensor("op_6912_dilations_0"), val = tensor([1, 1])]; + tensor var_6912_groups_0 = const()[name = tensor("op_6912_groups_0"), val = tensor(1)]; + tensor layers_17_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(410586496))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411372992))), name = tensor("layers_17_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6912_cast_fp16 = conv(dilations = var_6912_dilations_0, groups = var_6912_groups_0, pad = var_6912_pad_0, pad_type = var_6912_pad_type_0, strides = var_6912_strides_0, weight = layers_17_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_477_cast_fp16)[name = tensor("op_6912_cast_fp16")]; + tensor var_6918_pad_type_0 = const()[name = tensor("op_6918_pad_type_0"), val = tensor("valid")]; + tensor var_6918_strides_0 = const()[name = tensor("op_6918_strides_0"), val = tensor([1, 1])]; + tensor var_6918_pad_0 = const()[name = tensor("op_6918_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6918_dilations_0 = const()[name = tensor("op_6918_dilations_0"), val = tensor([1, 1])]; + tensor var_6918_groups_0 = const()[name = tensor("op_6918_groups_0"), val = tensor(1)]; + tensor layers_17_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411403904))), name = tensor("layers_17_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411373184))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_6918_cast_fp16 = conv(dilations = var_6918_dilations_0, groups = var_6918_groups_0, pad = var_6918_pad_0, pad_type = var_6918_pad_type_0, strides = var_6918_strides_0, weight = layers_17_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_477_cast_fp16)[name = tensor("op_6918_cast_fp16")]; + tensor x_107_cast_fp16 = add(x = var_6912_cast_fp16, y = var_6918_cast_fp16)[name = tensor("x_107_cast_fp16")]; + tensor inputs_177_cast_fp16 = add(x = inputs_175_cast_fp16, y = x_107_cast_fp16)[name = tensor("inputs_177_cast_fp16")]; + tensor out_177_axes_0 = const()[name = tensor("out_177_axes_0"), val = tensor([1])]; + tensor var_6929_to_fp16 = const()[name = tensor("op_6929_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_177_cast_fp16 = layer_norm(axes = out_177_axes_0, epsilon = var_6929_to_fp16, x = inputs_177_cast_fp16)[name = tensor("out_177_cast_fp16")]; + tensor input_479_gamma_0_to_fp16 = const()[name = tensor("input_479_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411535040)))]; + tensor input_479_beta_0_to_fp16 = const()[name = tensor("input_479_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411537152)))]; + tensor input_479_epsilon_0_to_fp16 = const()[name = tensor("input_479_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_479_cast_fp16 = batch_norm(beta = input_479_beta_0_to_fp16, epsilon = input_479_epsilon_0_to_fp16, gamma = input_479_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_177_cast_fp16)[name = tensor("input_479_cast_fp16")]; + tensor var_6949_pad_type_0 = const()[name = tensor("op_6949_pad_type_0"), val = tensor("valid")]; + tensor var_6949_strides_0 = const()[name = tensor("op_6949_strides_0"), val = tensor([1, 1])]; + tensor var_6949_pad_0 = const()[name = tensor("op_6949_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6949_dilations_0 = const()[name = tensor("op_6949_dilations_0"), val = tensor([1, 1])]; + tensor var_6949_groups_0 = const()[name = tensor("op_6949_groups_0"), val = tensor(1)]; + tensor layers_17_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411539264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(414685056))), name = tensor("layers_17_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_6949_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_6949_dilations_0, groups = var_6949_groups_0, pad = var_6949_pad_0, pad_type = var_6949_pad_type_0, strides = var_6949_strides_0, weight = layers_17_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_479_cast_fp16)[name = tensor("op_6949_cast_fp16")]; + tensor var_6955_pad_type_0 = const()[name = tensor("op_6955_pad_type_0"), val = tensor("valid")]; + tensor var_6955_strides_0 = const()[name = tensor("op_6955_strides_0"), val = tensor([1, 1])]; + tensor var_6955_pad_0 = const()[name = tensor("op_6955_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6955_dilations_0 = const()[name = tensor("op_6955_dilations_0"), val = tensor([1, 1])]; + tensor var_6955_groups_0 = const()[name = tensor("op_6955_groups_0"), val = tensor(1)]; + tensor layers_17_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(414819648))), name = tensor("layers_17_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(414685248))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_6955_cast_fp16 = conv(dilations = var_6955_dilations_0, groups = var_6955_groups_0, pad = var_6955_pad_0, pad_type = var_6955_pad_type_0, strides = var_6955_strides_0, weight = layers_17_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_479_cast_fp16)[name = tensor("op_6955_cast_fp16")]; + tensor input_481_cast_fp16 = add(x = var_6949_cast_fp16, y = var_6955_cast_fp16)[name = tensor("input_481_cast_fp16")]; + tensor input_483_cast_fp16 = silu(x = input_481_cast_fp16)[name = tensor("input_483_cast_fp16")]; + tensor var_6966_pad_type_0 = const()[name = tensor("op_6966_pad_type_0"), val = tensor("valid")]; + tensor var_6966_strides_0 = const()[name = tensor("op_6966_strides_0"), val = tensor([1, 1])]; + tensor var_6966_pad_0 = const()[name = tensor("op_6966_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6966_dilations_0 = const()[name = tensor("op_6966_dilations_0"), val = tensor([1, 1])]; + tensor var_6966_groups_0 = const()[name = tensor("op_6966_groups_0"), val = tensor(1)]; + tensor layers_17_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(415344000))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(418489792))), name = tensor("layers_17_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_6966_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6966_dilations_0, groups = var_6966_groups_0, pad = var_6966_pad_0, pad_type = var_6966_pad_type_0, strides = var_6966_strides_0, weight = layers_17_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_483_cast_fp16)[name = tensor("op_6966_cast_fp16")]; + tensor var_6972_pad_type_0 = const()[name = tensor("op_6972_pad_type_0"), val = tensor("valid")]; + tensor var_6972_strides_0 = const()[name = tensor("op_6972_strides_0"), val = tensor([1, 1])]; + tensor var_6972_pad_0 = const()[name = tensor("op_6972_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6972_dilations_0 = const()[name = tensor("op_6972_dilations_0"), val = tensor([1, 1])]; + tensor var_6972_groups_0 = const()[name = tensor("op_6972_groups_0"), val = tensor(1)]; + tensor layers_17_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(418631808))), name = tensor("layers_17_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(418489984))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_6972_cast_fp16 = conv(dilations = var_6972_dilations_0, groups = var_6972_groups_0, pad = var_6972_pad_0, pad_type = var_6972_pad_type_0, strides = var_6972_strides_0, weight = layers_17_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_483_cast_fp16)[name = tensor("op_6972_cast_fp16")]; + tensor x_109_cast_fp16 = add(x = var_6966_cast_fp16, y = var_6972_cast_fp16)[name = tensor("x_109_cast_fp16")]; + tensor var_6974_to_fp16 = const()[name = tensor("op_6974_to_fp16"), val = tensor(0x1p-1)]; + tensor var_6975_cast_fp16 = mul(x = x_109_cast_fp16, y = var_6974_to_fp16)[name = tensor("op_6975_cast_fp16")]; + tensor inputs_179_cast_fp16 = add(x = inputs_177_cast_fp16, y = var_6975_cast_fp16)[name = tensor("inputs_179_cast_fp16")]; + tensor out_179_axes_0 = const()[name = tensor("out_179_axes_0"), val = tensor([1])]; + tensor var_6985_to_fp16 = const()[name = tensor("op_6985_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_179_cast_fp16 = layer_norm(axes = out_179_axes_0, epsilon = var_6985_to_fp16, x = inputs_179_cast_fp16)[name = tensor("out_179_cast_fp16")]; + tensor inputs_181_gamma_0_to_fp16 = const()[name = tensor("inputs_181_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(419156160)))]; + tensor inputs_181_beta_0_to_fp16 = const()[name = tensor("inputs_181_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(419158272)))]; + tensor inputs_181_epsilon_0_to_fp16 = const()[name = tensor("inputs_181_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_181_cast_fp16 = batch_norm(beta = inputs_181_beta_0_to_fp16, epsilon = inputs_181_epsilon_0_to_fp16, gamma = inputs_181_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_179_cast_fp16)[name = tensor("inputs_181_cast_fp16")]; + tensor var_6999 = const()[name = tensor("op_6999"), val = tensor(3)]; + tensor out_181_axes_0 = const()[name = tensor("out_181_axes_0"), val = tensor([1])]; + tensor var_7030_to_fp16 = const()[name = tensor("op_7030_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_181_cast_fp16 = layer_norm(axes = out_181_axes_0, epsilon = var_7030_to_fp16, x = inputs_181_cast_fp16)[name = tensor("out_181_cast_fp16")]; + tensor input_485_gamma_0_to_fp16 = const()[name = tensor("input_485_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(419160384)))]; + tensor input_485_beta_0_to_fp16 = const()[name = tensor("input_485_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(419162496)))]; + tensor input_485_epsilon_0_to_fp16 = const()[name = tensor("input_485_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_485_cast_fp16 = batch_norm(beta = input_485_beta_0_to_fp16, epsilon = input_485_epsilon_0_to_fp16, gamma = input_485_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_181_cast_fp16)[name = tensor("input_485_cast_fp16")]; + tensor var_7050_pad_type_0 = const()[name = tensor("op_7050_pad_type_0"), val = tensor("valid")]; + tensor var_7050_strides_0 = const()[name = tensor("op_7050_strides_0"), val = tensor([1, 1])]; + tensor var_7050_pad_0 = const()[name = tensor("op_7050_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7050_dilations_0 = const()[name = tensor("op_7050_dilations_0"), val = tensor([1, 1])]; + tensor var_7050_groups_0 = const()[name = tensor("op_7050_groups_0"), val = tensor(1)]; + tensor layers_18_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(419164608))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(422310400))), name = tensor("layers_18_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_7050_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_7050_dilations_0, groups = var_7050_groups_0, pad = var_7050_pad_0, pad_type = var_7050_pad_type_0, strides = var_7050_strides_0, weight = layers_18_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_485_cast_fp16)[name = tensor("op_7050_cast_fp16")]; + tensor var_7056_pad_type_0 = const()[name = tensor("op_7056_pad_type_0"), val = tensor("valid")]; + tensor var_7056_strides_0 = const()[name = tensor("op_7056_strides_0"), val = tensor([1, 1])]; + tensor var_7056_pad_0 = const()[name = tensor("op_7056_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7056_dilations_0 = const()[name = tensor("op_7056_dilations_0"), val = tensor([1, 1])]; + tensor var_7056_groups_0 = const()[name = tensor("op_7056_groups_0"), val = tensor(1)]; + tensor layers_18_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(422450560))), name = tensor("layers_18_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(422310592))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_7056_cast_fp16 = conv(dilations = var_7056_dilations_0, groups = var_7056_groups_0, pad = var_7056_pad_0, pad_type = var_7056_pad_type_0, strides = var_7056_strides_0, weight = layers_18_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_485_cast_fp16)[name = tensor("op_7056_cast_fp16")]; + tensor input_487_cast_fp16 = add(x = var_7050_cast_fp16, y = var_7056_cast_fp16)[name = tensor("input_487_cast_fp16")]; + tensor input_489_cast_fp16 = silu(x = input_487_cast_fp16)[name = tensor("input_489_cast_fp16")]; + tensor var_7067_pad_type_0 = const()[name = tensor("op_7067_pad_type_0"), val = tensor("valid")]; + tensor var_7067_strides_0 = const()[name = tensor("op_7067_strides_0"), val = tensor([1, 1])]; + tensor var_7067_pad_0 = const()[name = tensor("op_7067_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7067_dilations_0 = const()[name = tensor("op_7067_dilations_0"), val = tensor([1, 1])]; + tensor var_7067_groups_0 = const()[name = tensor("op_7067_groups_0"), val = tensor(1)]; + tensor layers_18_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(422974912))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(426120704))), name = tensor("layers_18_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_7067_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7067_dilations_0, groups = var_7067_groups_0, pad = var_7067_pad_0, pad_type = var_7067_pad_type_0, strides = var_7067_strides_0, weight = layers_18_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_489_cast_fp16)[name = tensor("op_7067_cast_fp16")]; + tensor var_7073_pad_type_0 = const()[name = tensor("op_7073_pad_type_0"), val = tensor("valid")]; + tensor var_7073_strides_0 = const()[name = tensor("op_7073_strides_0"), val = tensor([1, 1])]; + tensor var_7073_pad_0 = const()[name = tensor("op_7073_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7073_dilations_0 = const()[name = tensor("op_7073_dilations_0"), val = tensor([1, 1])]; + tensor var_7073_groups_0 = const()[name = tensor("op_7073_groups_0"), val = tensor(1)]; + tensor layers_18_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(426259968))), name = tensor("layers_18_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(426120896))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_7073_cast_fp16 = conv(dilations = var_7073_dilations_0, groups = var_7073_groups_0, pad = var_7073_pad_0, pad_type = var_7073_pad_type_0, strides = var_7073_strides_0, weight = layers_18_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_489_cast_fp16)[name = tensor("op_7073_cast_fp16")]; + tensor x_111_cast_fp16 = add(x = var_7067_cast_fp16, y = var_7073_cast_fp16)[name = tensor("x_111_cast_fp16")]; + tensor var_7075_to_fp16 = const()[name = tensor("op_7075_to_fp16"), val = tensor(0x1p-1)]; + tensor var_7076_cast_fp16 = mul(x = x_111_cast_fp16, y = var_7075_to_fp16)[name = tensor("op_7076_cast_fp16")]; + tensor inputs_183_cast_fp16 = add(x = inputs_181_cast_fp16, y = var_7076_cast_fp16)[name = tensor("inputs_183_cast_fp16")]; + tensor out_183_axes_0 = const()[name = tensor("out_183_axes_0"), val = tensor([1])]; + tensor var_7086_to_fp16 = const()[name = tensor("op_7086_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_183_cast_fp16 = layer_norm(axes = out_183_axes_0, epsilon = var_7086_to_fp16, x = inputs_183_cast_fp16)[name = tensor("out_183_cast_fp16")]; + tensor obj_75_gamma_0_to_fp16 = const()[name = tensor("obj_75_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(426784320)))]; + tensor obj_75_beta_0_to_fp16 = const()[name = tensor("obj_75_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(426786432)))]; + tensor obj_75_epsilon_0_to_fp16 = const()[name = tensor("obj_75_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_75_cast_fp16 = batch_norm(beta = obj_75_beta_0_to_fp16, epsilon = obj_75_epsilon_0_to_fp16, gamma = obj_75_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_183_cast_fp16)[name = tensor("obj_75_cast_fp16")]; + tensor var_7111_pad_type_0 = const()[name = tensor("op_7111_pad_type_0"), val = tensor("valid")]; + tensor var_7111_strides_0 = const()[name = tensor("op_7111_strides_0"), val = tensor([1, 1])]; + tensor var_7111_pad_0 = const()[name = tensor("op_7111_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7111_dilations_0 = const()[name = tensor("op_7111_dilations_0"), val = tensor([1, 1])]; + tensor var_7111_groups_0 = const()[name = tensor("op_7111_groups_0"), val = tensor(1)]; + tensor layers_18_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(426788544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(427575040))), name = tensor("layers_18_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7111_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7111_dilations_0, groups = var_7111_groups_0, pad = var_7111_pad_0, pad_type = var_7111_pad_type_0, strides = var_7111_strides_0, weight = layers_18_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_75_cast_fp16)[name = tensor("op_7111_cast_fp16")]; + tensor var_7117_pad_type_0 = const()[name = tensor("op_7117_pad_type_0"), val = tensor("valid")]; + tensor var_7117_strides_0 = const()[name = tensor("op_7117_strides_0"), val = tensor([1, 1])]; + tensor var_7117_pad_0 = const()[name = tensor("op_7117_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7117_dilations_0 = const()[name = tensor("op_7117_dilations_0"), val = tensor([1, 1])]; + tensor var_7117_groups_0 = const()[name = tensor("op_7117_groups_0"), val = tensor(1)]; + tensor layers_18_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(427607488))), name = tensor("layers_18_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(427575232))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7117_cast_fp16 = conv(dilations = var_7117_dilations_0, groups = var_7117_groups_0, pad = var_7117_pad_0, pad_type = var_7117_pad_type_0, strides = var_7117_strides_0, weight = layers_18_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_75_cast_fp16)[name = tensor("op_7117_cast_fp16")]; + tensor query_73_cast_fp16 = add(x = var_7111_cast_fp16, y = var_7117_cast_fp16)[name = tensor("query_73_cast_fp16")]; + tensor var_7126_pad_type_0 = const()[name = tensor("op_7126_pad_type_0"), val = tensor("valid")]; + tensor var_7126_strides_0 = const()[name = tensor("op_7126_strides_0"), val = tensor([1, 1])]; + tensor var_7126_pad_0 = const()[name = tensor("op_7126_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7126_dilations_0 = const()[name = tensor("op_7126_dilations_0"), val = tensor([1, 1])]; + tensor var_7126_groups_0 = const()[name = tensor("op_7126_groups_0"), val = tensor(1)]; + tensor layers_18_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(427738624))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(428525120))), name = tensor("layers_18_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7126_cast_fp16 = conv(dilations = var_7126_dilations_0, groups = var_7126_groups_0, pad = var_7126_pad_0, pad_type = var_7126_pad_type_0, strides = var_7126_strides_0, weight = layers_18_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_75_cast_fp16)[name = tensor("op_7126_cast_fp16")]; + tensor var_7132_pad_type_0 = const()[name = tensor("op_7132_pad_type_0"), val = tensor("valid")]; + tensor var_7132_strides_0 = const()[name = tensor("op_7132_strides_0"), val = tensor([1, 1])]; + tensor var_7132_pad_0 = const()[name = tensor("op_7132_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7132_dilations_0 = const()[name = tensor("op_7132_dilations_0"), val = tensor([1, 1])]; + tensor var_7132_groups_0 = const()[name = tensor("op_7132_groups_0"), val = tensor(1)]; + tensor layers_18_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(428563712))), name = tensor("layers_18_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(428525312))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7132_cast_fp16 = conv(dilations = var_7132_dilations_0, groups = var_7132_groups_0, pad = var_7132_pad_0, pad_type = var_7132_pad_type_0, strides = var_7132_strides_0, weight = layers_18_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_75_cast_fp16)[name = tensor("op_7132_cast_fp16")]; + tensor key_37_cast_fp16 = add(x = var_7126_cast_fp16, y = var_7132_cast_fp16)[name = tensor("key_37_cast_fp16")]; + tensor var_7142_pad_type_0 = const()[name = tensor("op_7142_pad_type_0"), val = tensor("valid")]; + tensor var_7142_strides_0 = const()[name = tensor("op_7142_strides_0"), val = tensor([1, 1])]; + tensor var_7142_pad_0 = const()[name = tensor("op_7142_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7142_dilations_0 = const()[name = tensor("op_7142_dilations_0"), val = tensor([1, 1])]; + tensor var_7142_groups_0 = const()[name = tensor("op_7142_groups_0"), val = tensor(1)]; + tensor layers_18_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(428694848))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(429481344))), name = tensor("layers_18_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7142_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7142_dilations_0, groups = var_7142_groups_0, pad = var_7142_pad_0, pad_type = var_7142_pad_type_0, strides = var_7142_strides_0, weight = layers_18_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_75_cast_fp16)[name = tensor("op_7142_cast_fp16")]; + tensor var_7148_pad_type_0 = const()[name = tensor("op_7148_pad_type_0"), val = tensor("valid")]; + tensor var_7148_strides_0 = const()[name = tensor("op_7148_strides_0"), val = tensor([1, 1])]; + tensor var_7148_pad_0 = const()[name = tensor("op_7148_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7148_dilations_0 = const()[name = tensor("op_7148_dilations_0"), val = tensor([1, 1])]; + tensor var_7148_groups_0 = const()[name = tensor("op_7148_groups_0"), val = tensor(1)]; + tensor layers_18_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(429512768))), name = tensor("layers_18_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(429481536))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7148_cast_fp16 = conv(dilations = var_7148_dilations_0, groups = var_7148_groups_0, pad = var_7148_pad_0, pad_type = var_7148_pad_type_0, strides = var_7148_strides_0, weight = layers_18_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_75_cast_fp16)[name = tensor("op_7148_cast_fp16")]; + tensor value_37_cast_fp16 = add(x = var_7142_cast_fp16, y = var_7148_cast_fp16)[name = tensor("value_37_cast_fp16")]; + tensor var_7151_to_fp16 = const()[name = tensor("op_7151_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(429643904)))]; + tensor query_75_cast_fp16 = add(x = query_73_cast_fp16, y = var_7151_to_fp16)[name = tensor("query_75_cast_fp16")]; + tensor var_7154_to_fp16 = const()[name = tensor("op_7154_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(429646016)))]; + tensor q_with_bias_v_37_cast_fp16 = add(x = query_73_cast_fp16, y = var_7154_to_fp16)[name = tensor("q_with_bias_v_37_cast_fp16")]; + tensor var_7164_pad_type_0 = const()[name = tensor("op_7164_pad_type_0"), val = tensor("valid")]; + tensor var_7164_strides_0 = const()[name = tensor("op_7164_strides_0"), val = tensor([1, 1])]; + tensor var_7164_pad_0 = const()[name = tensor("op_7164_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7164_dilations_0 = const()[name = tensor("op_7164_dilations_0"), val = tensor([1, 1])]; + tensor var_7164_groups_0 = const()[name = tensor("op_7164_groups_0"), val = tensor(1)]; + tensor layers_18_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(429648128))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(430434624))), name = tensor("layers_18_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7164_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7164_dilations_0, groups = var_7164_groups_0, pad = var_7164_pad_0, pad_type = var_7164_pad_type_0, strides = var_7164_strides_0, weight = layers_18_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_7164_cast_fp16")]; + tensor var_7170_pad_type_0 = const()[name = tensor("op_7170_pad_type_0"), val = tensor("valid")]; + tensor var_7170_strides_0 = const()[name = tensor("op_7170_strides_0"), val = tensor([1, 1])]; + tensor var_7170_pad_0 = const()[name = tensor("op_7170_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7170_dilations_0 = const()[name = tensor("op_7170_dilations_0"), val = tensor([1, 1])]; + tensor var_7170_groups_0 = const()[name = tensor("op_7170_groups_0"), val = tensor(1)]; + tensor layers_18_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(430487104))), name = tensor("layers_18_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(430434816))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7170_cast_fp16 = conv(dilations = var_7170_dilations_0, groups = var_7170_groups_0, pad = var_7170_pad_0, pad_type = var_7170_pad_type_0, strides = var_7170_strides_0, weight = layers_18_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_7170_cast_fp16")]; + tensor p_37_cast_fp16 = add(x = var_7164_cast_fp16, y = var_7170_cast_fp16)[name = tensor("p_37_cast_fp16")]; + tensor var_7174 = const()[name = tensor("op_7174"), val = tensor([1, 8, 128, 188])]; + tensor var_7175_cast_fp16 = reshape(shape = var_7174, x = q_with_bias_v_37_cast_fp16)[name = tensor("op_7175_cast_fp16")]; + tensor var_7176 = const()[name = tensor("op_7176"), val = tensor([1, 8, 128, -1])]; + tensor var_7177_cast_fp16 = reshape(shape = var_7176, x = p_37_cast_fp16)[name = tensor("op_7177_cast_fp16")]; + tensor matrix_bd_145_transpose_x_0 = const()[name = tensor("matrix_bd_145_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_145_transpose_y_0 = const()[name = tensor("matrix_bd_145_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_145_cast_fp16 = matmul(transpose_x = matrix_bd_145_transpose_x_0, transpose_y = matrix_bd_145_transpose_y_0, x = var_7175_cast_fp16, y = var_7177_cast_fp16)[name = tensor("matrix_bd_145_cast_fp16")]; + tensor matrix_bd_147_pad_0 = const()[name = tensor("matrix_bd_147_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_147_mode_0 = const()[name = tensor("matrix_bd_147_mode_0"), val = tensor("constant")]; + tensor const_208_to_fp16 = const()[name = tensor("const_208_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_147_cast_fp16 = pad(constant_val = const_208_to_fp16, mode = matrix_bd_147_mode_0, pad = matrix_bd_147_pad_0, x = matrix_bd_145_cast_fp16)[name = tensor("matrix_bd_147_cast_fp16")]; + tensor var_7186 = const()[name = tensor("op_7186"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_149_cast_fp16 = reshape(shape = var_7186, x = matrix_bd_147_cast_fp16)[name = tensor("matrix_bd_149_cast_fp16")]; + tensor var_7190_begin_0 = const()[name = tensor("op_7190_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_7190_end_0 = const()[name = tensor("op_7190_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_7190_end_mask_0 = const()[name = tensor("op_7190_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_7190_cast_fp16 = slice_by_index(begin = var_7190_begin_0, end = var_7190_end_0, end_mask = var_7190_end_mask_0, x = matrix_bd_149_cast_fp16)[name = tensor("op_7190_cast_fp16")]; + tensor var_7191 = const()[name = tensor("op_7191"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_151_cast_fp16 = reshape(shape = var_7191, x = var_7190_cast_fp16)[name = tensor("matrix_bd_151_cast_fp16")]; + tensor var_7196_begin_0 = const()[name = tensor("op_7196_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7196_end_0 = const()[name = tensor("op_7196_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_7196_end_mask_0 = const()[name = tensor("op_7196_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7196_cast_fp16 = slice_by_index(begin = var_7196_begin_0, end = var_7196_end_0, end_mask = var_7196_end_mask_0, x = matrix_bd_151_cast_fp16)[name = tensor("op_7196_cast_fp16")]; + tensor var_7197_to_fp16 = const()[name = tensor("op_7197_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_37_cast_fp16 = mul(x = var_7196_cast_fp16, y = var_7197_to_fp16)[name = tensor("qk_mask_37_cast_fp16")]; + tensor var_7201 = const()[name = tensor("op_7201"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_37_cast_fp16 = reshape(shape = var_7201, x = query_75_cast_fp16)[name = tensor("mh_q_37_cast_fp16")]; + tensor var_7203_to_fp16 = const()[name = tensor("op_7203_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_7204_cast_fp16 = mul(x = mh_q_37_cast_fp16, y = var_7203_to_fp16)[name = tensor("op_7204_cast_fp16")]; + tensor var_7207 = const()[name = tensor("op_7207"), val = tensor([1, 8, 128, 188])]; + tensor var_7208_cast_fp16 = reshape(shape = var_7207, x = key_37_cast_fp16)[name = tensor("op_7208_cast_fp16")]; + tensor mh_w_73_transpose_x_0 = const()[name = tensor("mh_w_73_transpose_x_0"), val = tensor(true)]; + tensor mh_w_73_transpose_y_0 = const()[name = tensor("mh_w_73_transpose_y_0"), val = tensor(false)]; + tensor mh_w_73_cast_fp16 = matmul(transpose_x = mh_w_73_transpose_x_0, transpose_y = mh_w_73_transpose_y_0, x = var_7204_cast_fp16, y = var_7208_cast_fp16)[name = tensor("mh_w_73_cast_fp16")]; + tensor mh_w_75_cast_fp16 = add(x = mh_w_73_cast_fp16, y = qk_mask_37_cast_fp16)[name = tensor("mh_w_75_cast_fp16")]; + tensor var_7212_cast_fp16 = softmax(axis = var_6999, x = mh_w_75_cast_fp16)[name = tensor("op_7212_cast_fp16")]; + tensor var_7213 = const()[name = tensor("op_7213"), val = tensor([1, 8, 128, 188])]; + tensor var_7214_cast_fp16 = reshape(shape = var_7213, x = value_37_cast_fp16)[name = tensor("op_7214_cast_fp16")]; + tensor attn_37_transpose_x_0 = const()[name = tensor("attn_37_transpose_x_0"), val = tensor(false)]; + tensor attn_37_transpose_y_0 = const()[name = tensor("attn_37_transpose_y_0"), val = tensor(true)]; + tensor attn_37_cast_fp16 = matmul(transpose_x = attn_37_transpose_x_0, transpose_y = attn_37_transpose_y_0, x = var_7214_cast_fp16, y = var_7212_cast_fp16)[name = tensor("attn_37_cast_fp16")]; + tensor var_7217 = const()[name = tensor("op_7217"), val = tensor([1, 1024, 1, 188])]; + tensor input_491_cast_fp16 = reshape(shape = var_7217, x = attn_37_cast_fp16)[name = tensor("input_491_cast_fp16")]; + tensor var_7227_pad_type_0 = const()[name = tensor("op_7227_pad_type_0"), val = tensor("valid")]; + tensor var_7227_strides_0 = const()[name = tensor("op_7227_strides_0"), val = tensor([1, 1])]; + tensor var_7227_pad_0 = const()[name = tensor("op_7227_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7227_dilations_0 = const()[name = tensor("op_7227_dilations_0"), val = tensor([1, 1])]; + tensor var_7227_groups_0 = const()[name = tensor("op_7227_groups_0"), val = tensor(1)]; + tensor layers_18_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(430618240))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(431404736))), name = tensor("layers_18_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7227_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7227_dilations_0, groups = var_7227_groups_0, pad = var_7227_pad_0, pad_type = var_7227_pad_type_0, strides = var_7227_strides_0, weight = layers_18_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_491_cast_fp16)[name = tensor("op_7227_cast_fp16")]; + tensor var_7233_pad_type_0 = const()[name = tensor("op_7233_pad_type_0"), val = tensor("valid")]; + tensor var_7233_strides_0 = const()[name = tensor("op_7233_strides_0"), val = tensor([1, 1])]; + tensor var_7233_pad_0 = const()[name = tensor("op_7233_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7233_dilations_0 = const()[name = tensor("op_7233_dilations_0"), val = tensor([1, 1])]; + tensor var_7233_groups_0 = const()[name = tensor("op_7233_groups_0"), val = tensor(1)]; + tensor layers_18_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(431436992))), name = tensor("layers_18_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(431404928))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7233_cast_fp16 = conv(dilations = var_7233_dilations_0, groups = var_7233_groups_0, pad = var_7233_pad_0, pad_type = var_7233_pad_type_0, strides = var_7233_strides_0, weight = layers_18_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_491_cast_fp16)[name = tensor("op_7233_cast_fp16")]; + tensor obj_77_cast_fp16 = add(x = var_7227_cast_fp16, y = var_7233_cast_fp16)[name = tensor("obj_77_cast_fp16")]; + tensor inputs_185_cast_fp16 = add(x = inputs_183_cast_fp16, y = obj_77_cast_fp16)[name = tensor("inputs_185_cast_fp16")]; + tensor out_185_axes_0 = const()[name = tensor("out_185_axes_0"), val = tensor([1])]; + tensor var_7244_to_fp16 = const()[name = tensor("op_7244_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_185_cast_fp16 = layer_norm(axes = out_185_axes_0, epsilon = var_7244_to_fp16, x = inputs_185_cast_fp16)[name = tensor("out_185_cast_fp16")]; + tensor input_493_gamma_0_to_fp16 = const()[name = tensor("input_493_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(431568128)))]; + tensor input_493_beta_0_to_fp16 = const()[name = tensor("input_493_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(431570240)))]; + tensor input_493_epsilon_0_to_fp16 = const()[name = tensor("input_493_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_493_cast_fp16 = batch_norm(beta = input_493_beta_0_to_fp16, epsilon = input_493_epsilon_0_to_fp16, gamma = input_493_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_185_cast_fp16)[name = tensor("input_493_cast_fp16")]; + tensor var_7265_pad_type_0 = const()[name = tensor("op_7265_pad_type_0"), val = tensor("valid")]; + tensor var_7265_strides_0 = const()[name = tensor("op_7265_strides_0"), val = tensor([1, 1])]; + tensor var_7265_pad_0 = const()[name = tensor("op_7265_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7265_dilations_0 = const()[name = tensor("op_7265_dilations_0"), val = tensor([1, 1])]; + tensor var_7265_groups_0 = const()[name = tensor("op_7265_groups_0"), val = tensor(1)]; + tensor layers_18_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(431572352))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(433145280))), name = tensor("layers_18_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_7265_cast_fp16 = conv(dilations = var_7265_dilations_0, groups = var_7265_groups_0, pad = var_7265_pad_0, pad_type = var_7265_pad_type_0, strides = var_7265_strides_0, weight = layers_18_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_493_cast_fp16)[name = tensor("op_7265_cast_fp16")]; + tensor var_7271_pad_type_0 = const()[name = tensor("op_7271_pad_type_0"), val = tensor("valid")]; + tensor var_7271_strides_0 = const()[name = tensor("op_7271_strides_0"), val = tensor([1, 1])]; + tensor var_7271_pad_0 = const()[name = tensor("op_7271_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7271_dilations_0 = const()[name = tensor("op_7271_dilations_0"), val = tensor([1, 1])]; + tensor var_7271_groups_0 = const()[name = tensor("op_7271_groups_0"), val = tensor(1)]; + tensor layers_18_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(433212544))), name = tensor("layers_18_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(433145472))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_7271_cast_fp16 = conv(dilations = var_7271_dilations_0, groups = var_7271_groups_0, pad = var_7271_pad_0, pad_type = var_7271_pad_type_0, strides = var_7271_strides_0, weight = layers_18_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_493_cast_fp16)[name = tensor("op_7271_cast_fp16")]; + tensor input_495_cast_fp16 = add(x = var_7265_cast_fp16, y = var_7271_cast_fp16)[name = tensor("input_495_cast_fp16")]; + tensor input_497_split_num_splits_0 = const()[name = tensor("input_497_split_num_splits_0"), val = tensor(2)]; + tensor input_497_split_axis_0 = const()[name = tensor("input_497_split_axis_0"), val = tensor(1)]; + tensor input_497_split_cast_fp16_0, tensor input_497_split_cast_fp16_1 = split(axis = input_497_split_axis_0, num_splits = input_497_split_num_splits_0, x = input_495_cast_fp16)[name = tensor("input_497_split_cast_fp16")]; + tensor input_497_split_1_sigmoid_cast_fp16 = sigmoid(x = input_497_split_cast_fp16_1)[name = tensor("input_497_split_1_sigmoid_cast_fp16")]; + tensor input_497_cast_fp16 = mul(x = input_497_split_cast_fp16_0, y = input_497_split_1_sigmoid_cast_fp16)[name = tensor("input_497_cast_fp16")]; + tensor input_499_pad_type_0 = const()[name = tensor("input_499_pad_type_0"), val = tensor("custom")]; + tensor input_499_pad_0 = const()[name = tensor("input_499_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_499_groups_0 = const()[name = tensor("input_499_groups_0"), val = tensor(1024)]; + tensor input_499_strides_0 = const()[name = tensor("input_499_strides_0"), val = tensor([1, 1])]; + tensor input_499_dilations_0 = const()[name = tensor("input_499_dilations_0"), val = tensor([1, 1])]; + tensor const_304_to_fp16 = const()[name = tensor("const_304_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(433474752)))]; + tensor const_305_to_fp16 = const()[name = tensor("const_305_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(433493248)))]; + tensor input_501_cast_fp16 = conv(bias = const_305_to_fp16, dilations = input_499_dilations_0, groups = input_499_groups_0, pad = input_499_pad_0, pad_type = input_499_pad_type_0, strides = input_499_strides_0, weight = const_304_to_fp16, x = input_497_cast_fp16)[name = tensor("input_501_cast_fp16")]; + tensor input_503_cast_fp16 = silu(x = input_501_cast_fp16)[name = tensor("input_503_cast_fp16")]; + tensor var_7293_pad_type_0 = const()[name = tensor("op_7293_pad_type_0"), val = tensor("valid")]; + tensor var_7293_strides_0 = const()[name = tensor("op_7293_strides_0"), val = tensor([1, 1])]; + tensor var_7293_pad_0 = const()[name = tensor("op_7293_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7293_dilations_0 = const()[name = tensor("op_7293_dilations_0"), val = tensor([1, 1])]; + tensor var_7293_groups_0 = const()[name = tensor("op_7293_groups_0"), val = tensor(1)]; + tensor layers_18_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(433495360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434281856))), name = tensor("layers_18_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7293_cast_fp16 = conv(dilations = var_7293_dilations_0, groups = var_7293_groups_0, pad = var_7293_pad_0, pad_type = var_7293_pad_type_0, strides = var_7293_strides_0, weight = layers_18_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_503_cast_fp16)[name = tensor("op_7293_cast_fp16")]; + tensor var_7299_pad_type_0 = const()[name = tensor("op_7299_pad_type_0"), val = tensor("valid")]; + tensor var_7299_strides_0 = const()[name = tensor("op_7299_strides_0"), val = tensor([1, 1])]; + tensor var_7299_pad_0 = const()[name = tensor("op_7299_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7299_dilations_0 = const()[name = tensor("op_7299_dilations_0"), val = tensor([1, 1])]; + tensor var_7299_groups_0 = const()[name = tensor("op_7299_groups_0"), val = tensor(1)]; + tensor layers_18_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434312640))), name = tensor("layers_18_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434282048))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7299_cast_fp16 = conv(dilations = var_7299_dilations_0, groups = var_7299_groups_0, pad = var_7299_pad_0, pad_type = var_7299_pad_type_0, strides = var_7299_strides_0, weight = layers_18_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_503_cast_fp16)[name = tensor("op_7299_cast_fp16")]; + tensor x_113_cast_fp16 = add(x = var_7293_cast_fp16, y = var_7299_cast_fp16)[name = tensor("x_113_cast_fp16")]; + tensor inputs_187_cast_fp16 = add(x = inputs_185_cast_fp16, y = x_113_cast_fp16)[name = tensor("inputs_187_cast_fp16")]; + tensor out_187_axes_0 = const()[name = tensor("out_187_axes_0"), val = tensor([1])]; + tensor var_7310_to_fp16 = const()[name = tensor("op_7310_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_187_cast_fp16 = layer_norm(axes = out_187_axes_0, epsilon = var_7310_to_fp16, x = inputs_187_cast_fp16)[name = tensor("out_187_cast_fp16")]; + tensor input_505_gamma_0_to_fp16 = const()[name = tensor("input_505_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434443776)))]; + tensor input_505_beta_0_to_fp16 = const()[name = tensor("input_505_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434445888)))]; + tensor input_505_epsilon_0_to_fp16 = const()[name = tensor("input_505_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_505_cast_fp16 = batch_norm(beta = input_505_beta_0_to_fp16, epsilon = input_505_epsilon_0_to_fp16, gamma = input_505_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_187_cast_fp16)[name = tensor("input_505_cast_fp16")]; + tensor var_7330_pad_type_0 = const()[name = tensor("op_7330_pad_type_0"), val = tensor("valid")]; + tensor var_7330_strides_0 = const()[name = tensor("op_7330_strides_0"), val = tensor([1, 1])]; + tensor var_7330_pad_0 = const()[name = tensor("op_7330_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7330_dilations_0 = const()[name = tensor("op_7330_dilations_0"), val = tensor([1, 1])]; + tensor var_7330_groups_0 = const()[name = tensor("op_7330_groups_0"), val = tensor(1)]; + tensor layers_18_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434448000))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(437593792))), name = tensor("layers_18_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_7330_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_7330_dilations_0, groups = var_7330_groups_0, pad = var_7330_pad_0, pad_type = var_7330_pad_type_0, strides = var_7330_strides_0, weight = layers_18_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_505_cast_fp16)[name = tensor("op_7330_cast_fp16")]; + tensor var_7336_pad_type_0 = const()[name = tensor("op_7336_pad_type_0"), val = tensor("valid")]; + tensor var_7336_strides_0 = const()[name = tensor("op_7336_strides_0"), val = tensor([1, 1])]; + tensor var_7336_pad_0 = const()[name = tensor("op_7336_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7336_dilations_0 = const()[name = tensor("op_7336_dilations_0"), val = tensor([1, 1])]; + tensor var_7336_groups_0 = const()[name = tensor("op_7336_groups_0"), val = tensor(1)]; + tensor layers_18_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(437725376))), name = tensor("layers_18_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(437593984))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_7336_cast_fp16 = conv(dilations = var_7336_dilations_0, groups = var_7336_groups_0, pad = var_7336_pad_0, pad_type = var_7336_pad_type_0, strides = var_7336_strides_0, weight = layers_18_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_505_cast_fp16)[name = tensor("op_7336_cast_fp16")]; + tensor input_507_cast_fp16 = add(x = var_7330_cast_fp16, y = var_7336_cast_fp16)[name = tensor("input_507_cast_fp16")]; + tensor input_509_cast_fp16 = silu(x = input_507_cast_fp16)[name = tensor("input_509_cast_fp16")]; + tensor var_7347_pad_type_0 = const()[name = tensor("op_7347_pad_type_0"), val = tensor("valid")]; + tensor var_7347_strides_0 = const()[name = tensor("op_7347_strides_0"), val = tensor([1, 1])]; + tensor var_7347_pad_0 = const()[name = tensor("op_7347_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7347_dilations_0 = const()[name = tensor("op_7347_dilations_0"), val = tensor([1, 1])]; + tensor var_7347_groups_0 = const()[name = tensor("op_7347_groups_0"), val = tensor(1)]; + tensor layers_18_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(438249728))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(441395520))), name = tensor("layers_18_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_7347_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7347_dilations_0, groups = var_7347_groups_0, pad = var_7347_pad_0, pad_type = var_7347_pad_type_0, strides = var_7347_strides_0, weight = layers_18_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_509_cast_fp16)[name = tensor("op_7347_cast_fp16")]; + tensor var_7353_pad_type_0 = const()[name = tensor("op_7353_pad_type_0"), val = tensor("valid")]; + tensor var_7353_strides_0 = const()[name = tensor("op_7353_strides_0"), val = tensor([1, 1])]; + tensor var_7353_pad_0 = const()[name = tensor("op_7353_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7353_dilations_0 = const()[name = tensor("op_7353_dilations_0"), val = tensor([1, 1])]; + tensor var_7353_groups_0 = const()[name = tensor("op_7353_groups_0"), val = tensor(1)]; + tensor layers_18_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(441533568))), name = tensor("layers_18_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(441395712))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_7353_cast_fp16 = conv(dilations = var_7353_dilations_0, groups = var_7353_groups_0, pad = var_7353_pad_0, pad_type = var_7353_pad_type_0, strides = var_7353_strides_0, weight = layers_18_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_509_cast_fp16)[name = tensor("op_7353_cast_fp16")]; + tensor x_115_cast_fp16 = add(x = var_7347_cast_fp16, y = var_7353_cast_fp16)[name = tensor("x_115_cast_fp16")]; + tensor var_7355_to_fp16 = const()[name = tensor("op_7355_to_fp16"), val = tensor(0x1p-1)]; + tensor var_7356_cast_fp16 = mul(x = x_115_cast_fp16, y = var_7355_to_fp16)[name = tensor("op_7356_cast_fp16")]; + tensor inputs_189_cast_fp16 = add(x = inputs_187_cast_fp16, y = var_7356_cast_fp16)[name = tensor("inputs_189_cast_fp16")]; + tensor out_189_axes_0 = const()[name = tensor("out_189_axes_0"), val = tensor([1])]; + tensor var_7366_to_fp16 = const()[name = tensor("op_7366_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_189_cast_fp16 = layer_norm(axes = out_189_axes_0, epsilon = var_7366_to_fp16, x = inputs_189_cast_fp16)[name = tensor("out_189_cast_fp16")]; + tensor inputs_191_gamma_0_to_fp16 = const()[name = tensor("inputs_191_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(442057920)))]; + tensor inputs_191_beta_0_to_fp16 = const()[name = tensor("inputs_191_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(442060032)))]; + tensor inputs_191_epsilon_0_to_fp16 = const()[name = tensor("inputs_191_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_191_cast_fp16 = batch_norm(beta = inputs_191_beta_0_to_fp16, epsilon = inputs_191_epsilon_0_to_fp16, gamma = inputs_191_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_189_cast_fp16)[name = tensor("inputs_191_cast_fp16")]; + tensor var_7380 = const()[name = tensor("op_7380"), val = tensor(3)]; + tensor out_191_axes_0 = const()[name = tensor("out_191_axes_0"), val = tensor([1])]; + tensor var_7411_to_fp16 = const()[name = tensor("op_7411_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_191_cast_fp16 = layer_norm(axes = out_191_axes_0, epsilon = var_7411_to_fp16, x = inputs_191_cast_fp16)[name = tensor("out_191_cast_fp16")]; + tensor input_511_gamma_0_to_fp16 = const()[name = tensor("input_511_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(442062144)))]; + tensor input_511_beta_0_to_fp16 = const()[name = tensor("input_511_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(442064256)))]; + tensor input_511_epsilon_0_to_fp16 = const()[name = tensor("input_511_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_511_cast_fp16 = batch_norm(beta = input_511_beta_0_to_fp16, epsilon = input_511_epsilon_0_to_fp16, gamma = input_511_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_191_cast_fp16)[name = tensor("input_511_cast_fp16")]; + tensor var_7431_pad_type_0 = const()[name = tensor("op_7431_pad_type_0"), val = tensor("valid")]; + tensor var_7431_strides_0 = const()[name = tensor("op_7431_strides_0"), val = tensor([1, 1])]; + tensor var_7431_pad_0 = const()[name = tensor("op_7431_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7431_dilations_0 = const()[name = tensor("op_7431_dilations_0"), val = tensor([1, 1])]; + tensor var_7431_groups_0 = const()[name = tensor("op_7431_groups_0"), val = tensor(1)]; + tensor layers_19_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(442066368))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(445212160))), name = tensor("layers_19_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_7431_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_7431_dilations_0, groups = var_7431_groups_0, pad = var_7431_pad_0, pad_type = var_7431_pad_type_0, strides = var_7431_strides_0, weight = layers_19_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_511_cast_fp16)[name = tensor("op_7431_cast_fp16")]; + tensor var_7437_pad_type_0 = const()[name = tensor("op_7437_pad_type_0"), val = tensor("valid")]; + tensor var_7437_strides_0 = const()[name = tensor("op_7437_strides_0"), val = tensor([1, 1])]; + tensor var_7437_pad_0 = const()[name = tensor("op_7437_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7437_dilations_0 = const()[name = tensor("op_7437_dilations_0"), val = tensor([1, 1])]; + tensor var_7437_groups_0 = const()[name = tensor("op_7437_groups_0"), val = tensor(1)]; + tensor layers_19_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(445345600))), name = tensor("layers_19_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(445212352))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_7437_cast_fp16 = conv(dilations = var_7437_dilations_0, groups = var_7437_groups_0, pad = var_7437_pad_0, pad_type = var_7437_pad_type_0, strides = var_7437_strides_0, weight = layers_19_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_511_cast_fp16)[name = tensor("op_7437_cast_fp16")]; + tensor input_513_cast_fp16 = add(x = var_7431_cast_fp16, y = var_7437_cast_fp16)[name = tensor("input_513_cast_fp16")]; + tensor input_515_cast_fp16 = silu(x = input_513_cast_fp16)[name = tensor("input_515_cast_fp16")]; + tensor var_7448_pad_type_0 = const()[name = tensor("op_7448_pad_type_0"), val = tensor("valid")]; + tensor var_7448_strides_0 = const()[name = tensor("op_7448_strides_0"), val = tensor([1, 1])]; + tensor var_7448_pad_0 = const()[name = tensor("op_7448_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7448_dilations_0 = const()[name = tensor("op_7448_dilations_0"), val = tensor([1, 1])]; + tensor var_7448_groups_0 = const()[name = tensor("op_7448_groups_0"), val = tensor(1)]; + tensor layers_19_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(445869952))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(449015744))), name = tensor("layers_19_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_7448_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7448_dilations_0, groups = var_7448_groups_0, pad = var_7448_pad_0, pad_type = var_7448_pad_type_0, strides = var_7448_strides_0, weight = layers_19_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_515_cast_fp16)[name = tensor("op_7448_cast_fp16")]; + tensor var_7454_pad_type_0 = const()[name = tensor("op_7454_pad_type_0"), val = tensor("valid")]; + tensor var_7454_strides_0 = const()[name = tensor("op_7454_strides_0"), val = tensor([1, 1])]; + tensor var_7454_pad_0 = const()[name = tensor("op_7454_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7454_dilations_0 = const()[name = tensor("op_7454_dilations_0"), val = tensor([1, 1])]; + tensor var_7454_groups_0 = const()[name = tensor("op_7454_groups_0"), val = tensor(1)]; + tensor layers_19_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(449155520))), name = tensor("layers_19_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(449015936))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_7454_cast_fp16 = conv(dilations = var_7454_dilations_0, groups = var_7454_groups_0, pad = var_7454_pad_0, pad_type = var_7454_pad_type_0, strides = var_7454_strides_0, weight = layers_19_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_515_cast_fp16)[name = tensor("op_7454_cast_fp16")]; + tensor x_117_cast_fp16 = add(x = var_7448_cast_fp16, y = var_7454_cast_fp16)[name = tensor("x_117_cast_fp16")]; + tensor var_7456_to_fp16 = const()[name = tensor("op_7456_to_fp16"), val = tensor(0x1p-1)]; + tensor var_7457_cast_fp16 = mul(x = x_117_cast_fp16, y = var_7456_to_fp16)[name = tensor("op_7457_cast_fp16")]; + tensor inputs_193_cast_fp16 = add(x = inputs_191_cast_fp16, y = var_7457_cast_fp16)[name = tensor("inputs_193_cast_fp16")]; + tensor out_193_axes_0 = const()[name = tensor("out_193_axes_0"), val = tensor([1])]; + tensor var_7467_to_fp16 = const()[name = tensor("op_7467_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_193_cast_fp16 = layer_norm(axes = out_193_axes_0, epsilon = var_7467_to_fp16, x = inputs_193_cast_fp16)[name = tensor("out_193_cast_fp16")]; + tensor obj_79_gamma_0_to_fp16 = const()[name = tensor("obj_79_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(449679872)))]; + tensor obj_79_beta_0_to_fp16 = const()[name = tensor("obj_79_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(449681984)))]; + tensor obj_79_epsilon_0_to_fp16 = const()[name = tensor("obj_79_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_79_cast_fp16 = batch_norm(beta = obj_79_beta_0_to_fp16, epsilon = obj_79_epsilon_0_to_fp16, gamma = obj_79_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_193_cast_fp16)[name = tensor("obj_79_cast_fp16")]; + tensor var_7492_pad_type_0 = const()[name = tensor("op_7492_pad_type_0"), val = tensor("valid")]; + tensor var_7492_strides_0 = const()[name = tensor("op_7492_strides_0"), val = tensor([1, 1])]; + tensor var_7492_pad_0 = const()[name = tensor("op_7492_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7492_dilations_0 = const()[name = tensor("op_7492_dilations_0"), val = tensor([1, 1])]; + tensor var_7492_groups_0 = const()[name = tensor("op_7492_groups_0"), val = tensor(1)]; + tensor layers_19_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(449684096))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450470592))), name = tensor("layers_19_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7492_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7492_dilations_0, groups = var_7492_groups_0, pad = var_7492_pad_0, pad_type = var_7492_pad_type_0, strides = var_7492_strides_0, weight = layers_19_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_79_cast_fp16)[name = tensor("op_7492_cast_fp16")]; + tensor var_7498_pad_type_0 = const()[name = tensor("op_7498_pad_type_0"), val = tensor("valid")]; + tensor var_7498_strides_0 = const()[name = tensor("op_7498_strides_0"), val = tensor([1, 1])]; + tensor var_7498_pad_0 = const()[name = tensor("op_7498_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7498_dilations_0 = const()[name = tensor("op_7498_dilations_0"), val = tensor([1, 1])]; + tensor var_7498_groups_0 = const()[name = tensor("op_7498_groups_0"), val = tensor(1)]; + tensor layers_19_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450504128))), name = tensor("layers_19_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450470784))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7498_cast_fp16 = conv(dilations = var_7498_dilations_0, groups = var_7498_groups_0, pad = var_7498_pad_0, pad_type = var_7498_pad_type_0, strides = var_7498_strides_0, weight = layers_19_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_79_cast_fp16)[name = tensor("op_7498_cast_fp16")]; + tensor query_77_cast_fp16 = add(x = var_7492_cast_fp16, y = var_7498_cast_fp16)[name = tensor("query_77_cast_fp16")]; + tensor var_7507_pad_type_0 = const()[name = tensor("op_7507_pad_type_0"), val = tensor("valid")]; + tensor var_7507_strides_0 = const()[name = tensor("op_7507_strides_0"), val = tensor([1, 1])]; + tensor var_7507_pad_0 = const()[name = tensor("op_7507_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7507_dilations_0 = const()[name = tensor("op_7507_dilations_0"), val = tensor([1, 1])]; + tensor var_7507_groups_0 = const()[name = tensor("op_7507_groups_0"), val = tensor(1)]; + tensor layers_19_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450635264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(451421760))), name = tensor("layers_19_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7507_cast_fp16 = conv(dilations = var_7507_dilations_0, groups = var_7507_groups_0, pad = var_7507_pad_0, pad_type = var_7507_pad_type_0, strides = var_7507_strides_0, weight = layers_19_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_79_cast_fp16)[name = tensor("op_7507_cast_fp16")]; + tensor var_7513_pad_type_0 = const()[name = tensor("op_7513_pad_type_0"), val = tensor("valid")]; + tensor var_7513_strides_0 = const()[name = tensor("op_7513_strides_0"), val = tensor([1, 1])]; + tensor var_7513_pad_0 = const()[name = tensor("op_7513_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7513_dilations_0 = const()[name = tensor("op_7513_dilations_0"), val = tensor([1, 1])]; + tensor var_7513_groups_0 = const()[name = tensor("op_7513_groups_0"), val = tensor(1)]; + tensor layers_19_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(451455552))), name = tensor("layers_19_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(451421952))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7513_cast_fp16 = conv(dilations = var_7513_dilations_0, groups = var_7513_groups_0, pad = var_7513_pad_0, pad_type = var_7513_pad_type_0, strides = var_7513_strides_0, weight = layers_19_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_79_cast_fp16)[name = tensor("op_7513_cast_fp16")]; + tensor key_39_cast_fp16 = add(x = var_7507_cast_fp16, y = var_7513_cast_fp16)[name = tensor("key_39_cast_fp16")]; + tensor var_7523_pad_type_0 = const()[name = tensor("op_7523_pad_type_0"), val = tensor("valid")]; + tensor var_7523_strides_0 = const()[name = tensor("op_7523_strides_0"), val = tensor([1, 1])]; + tensor var_7523_pad_0 = const()[name = tensor("op_7523_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7523_dilations_0 = const()[name = tensor("op_7523_dilations_0"), val = tensor([1, 1])]; + tensor var_7523_groups_0 = const()[name = tensor("op_7523_groups_0"), val = tensor(1)]; + tensor layers_19_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(451586688))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(452373184))), name = tensor("layers_19_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7523_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7523_dilations_0, groups = var_7523_groups_0, pad = var_7523_pad_0, pad_type = var_7523_pad_type_0, strides = var_7523_strides_0, weight = layers_19_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_79_cast_fp16)[name = tensor("op_7523_cast_fp16")]; + tensor var_7529_pad_type_0 = const()[name = tensor("op_7529_pad_type_0"), val = tensor("valid")]; + tensor var_7529_strides_0 = const()[name = tensor("op_7529_strides_0"), val = tensor([1, 1])]; + tensor var_7529_pad_0 = const()[name = tensor("op_7529_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7529_dilations_0 = const()[name = tensor("op_7529_dilations_0"), val = tensor([1, 1])]; + tensor var_7529_groups_0 = const()[name = tensor("op_7529_groups_0"), val = tensor(1)]; + tensor layers_19_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(452402304))), name = tensor("layers_19_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(452373376))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7529_cast_fp16 = conv(dilations = var_7529_dilations_0, groups = var_7529_groups_0, pad = var_7529_pad_0, pad_type = var_7529_pad_type_0, strides = var_7529_strides_0, weight = layers_19_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_79_cast_fp16)[name = tensor("op_7529_cast_fp16")]; + tensor value_39_cast_fp16 = add(x = var_7523_cast_fp16, y = var_7529_cast_fp16)[name = tensor("value_39_cast_fp16")]; + tensor var_7532_to_fp16 = const()[name = tensor("op_7532_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(452533440)))]; + tensor query_79_cast_fp16 = add(x = query_77_cast_fp16, y = var_7532_to_fp16)[name = tensor("query_79_cast_fp16")]; + tensor var_7535_to_fp16 = const()[name = tensor("op_7535_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(452535552)))]; + tensor q_with_bias_v_39_cast_fp16 = add(x = query_77_cast_fp16, y = var_7535_to_fp16)[name = tensor("q_with_bias_v_39_cast_fp16")]; + tensor var_7545_pad_type_0 = const()[name = tensor("op_7545_pad_type_0"), val = tensor("valid")]; + tensor var_7545_strides_0 = const()[name = tensor("op_7545_strides_0"), val = tensor([1, 1])]; + tensor var_7545_pad_0 = const()[name = tensor("op_7545_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7545_dilations_0 = const()[name = tensor("op_7545_dilations_0"), val = tensor([1, 1])]; + tensor var_7545_groups_0 = const()[name = tensor("op_7545_groups_0"), val = tensor(1)]; + tensor layers_19_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(452537664))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(453324160))), name = tensor("layers_19_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7545_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7545_dilations_0, groups = var_7545_groups_0, pad = var_7545_pad_0, pad_type = var_7545_pad_type_0, strides = var_7545_strides_0, weight = layers_19_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_7545_cast_fp16")]; + tensor var_7551_pad_type_0 = const()[name = tensor("op_7551_pad_type_0"), val = tensor("valid")]; + tensor var_7551_strides_0 = const()[name = tensor("op_7551_strides_0"), val = tensor([1, 1])]; + tensor var_7551_pad_0 = const()[name = tensor("op_7551_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7551_dilations_0 = const()[name = tensor("op_7551_dilations_0"), val = tensor([1, 1])]; + tensor var_7551_groups_0 = const()[name = tensor("op_7551_groups_0"), val = tensor(1)]; + tensor layers_19_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(453382336))), name = tensor("layers_19_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(453324352))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7551_cast_fp16 = conv(dilations = var_7551_dilations_0, groups = var_7551_groups_0, pad = var_7551_pad_0, pad_type = var_7551_pad_type_0, strides = var_7551_strides_0, weight = layers_19_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_7551_cast_fp16")]; + tensor p_39_cast_fp16 = add(x = var_7545_cast_fp16, y = var_7551_cast_fp16)[name = tensor("p_39_cast_fp16")]; + tensor var_7555 = const()[name = tensor("op_7555"), val = tensor([1, 8, 128, 188])]; + tensor var_7556_cast_fp16 = reshape(shape = var_7555, x = q_with_bias_v_39_cast_fp16)[name = tensor("op_7556_cast_fp16")]; + tensor var_7557 = const()[name = tensor("op_7557"), val = tensor([1, 8, 128, -1])]; + tensor var_7558_cast_fp16 = reshape(shape = var_7557, x = p_39_cast_fp16)[name = tensor("op_7558_cast_fp16")]; + tensor matrix_bd_153_transpose_x_0 = const()[name = tensor("matrix_bd_153_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_153_transpose_y_0 = const()[name = tensor("matrix_bd_153_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_153_cast_fp16 = matmul(transpose_x = matrix_bd_153_transpose_x_0, transpose_y = matrix_bd_153_transpose_y_0, x = var_7556_cast_fp16, y = var_7558_cast_fp16)[name = tensor("matrix_bd_153_cast_fp16")]; + tensor matrix_bd_155_pad_0 = const()[name = tensor("matrix_bd_155_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_155_mode_0 = const()[name = tensor("matrix_bd_155_mode_0"), val = tensor("constant")]; + tensor const_219_to_fp16 = const()[name = tensor("const_219_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_155_cast_fp16 = pad(constant_val = const_219_to_fp16, mode = matrix_bd_155_mode_0, pad = matrix_bd_155_pad_0, x = matrix_bd_153_cast_fp16)[name = tensor("matrix_bd_155_cast_fp16")]; + tensor var_7567 = const()[name = tensor("op_7567"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_157_cast_fp16 = reshape(shape = var_7567, x = matrix_bd_155_cast_fp16)[name = tensor("matrix_bd_157_cast_fp16")]; + tensor var_7571_begin_0 = const()[name = tensor("op_7571_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_7571_end_0 = const()[name = tensor("op_7571_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_7571_end_mask_0 = const()[name = tensor("op_7571_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_7571_cast_fp16 = slice_by_index(begin = var_7571_begin_0, end = var_7571_end_0, end_mask = var_7571_end_mask_0, x = matrix_bd_157_cast_fp16)[name = tensor("op_7571_cast_fp16")]; + tensor var_7572 = const()[name = tensor("op_7572"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_159_cast_fp16 = reshape(shape = var_7572, x = var_7571_cast_fp16)[name = tensor("matrix_bd_159_cast_fp16")]; + tensor var_7577_begin_0 = const()[name = tensor("op_7577_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7577_end_0 = const()[name = tensor("op_7577_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_7577_end_mask_0 = const()[name = tensor("op_7577_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7577_cast_fp16 = slice_by_index(begin = var_7577_begin_0, end = var_7577_end_0, end_mask = var_7577_end_mask_0, x = matrix_bd_159_cast_fp16)[name = tensor("op_7577_cast_fp16")]; + tensor var_7578_to_fp16 = const()[name = tensor("op_7578_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_39_cast_fp16 = mul(x = var_7577_cast_fp16, y = var_7578_to_fp16)[name = tensor("qk_mask_39_cast_fp16")]; + tensor var_7582 = const()[name = tensor("op_7582"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_39_cast_fp16 = reshape(shape = var_7582, x = query_79_cast_fp16)[name = tensor("mh_q_39_cast_fp16")]; + tensor var_7584_to_fp16 = const()[name = tensor("op_7584_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_7585_cast_fp16 = mul(x = mh_q_39_cast_fp16, y = var_7584_to_fp16)[name = tensor("op_7585_cast_fp16")]; + tensor var_7588 = const()[name = tensor("op_7588"), val = tensor([1, 8, 128, 188])]; + tensor var_7589_cast_fp16 = reshape(shape = var_7588, x = key_39_cast_fp16)[name = tensor("op_7589_cast_fp16")]; + tensor mh_w_77_transpose_x_0 = const()[name = tensor("mh_w_77_transpose_x_0"), val = tensor(true)]; + tensor mh_w_77_transpose_y_0 = const()[name = tensor("mh_w_77_transpose_y_0"), val = tensor(false)]; + tensor mh_w_77_cast_fp16 = matmul(transpose_x = mh_w_77_transpose_x_0, transpose_y = mh_w_77_transpose_y_0, x = var_7585_cast_fp16, y = var_7589_cast_fp16)[name = tensor("mh_w_77_cast_fp16")]; + tensor mh_w_79_cast_fp16 = add(x = mh_w_77_cast_fp16, y = qk_mask_39_cast_fp16)[name = tensor("mh_w_79_cast_fp16")]; + tensor var_7593_cast_fp16 = softmax(axis = var_7380, x = mh_w_79_cast_fp16)[name = tensor("op_7593_cast_fp16")]; + tensor var_7594 = const()[name = tensor("op_7594"), val = tensor([1, 8, 128, 188])]; + tensor var_7595_cast_fp16 = reshape(shape = var_7594, x = value_39_cast_fp16)[name = tensor("op_7595_cast_fp16")]; + tensor attn_39_transpose_x_0 = const()[name = tensor("attn_39_transpose_x_0"), val = tensor(false)]; + tensor attn_39_transpose_y_0 = const()[name = tensor("attn_39_transpose_y_0"), val = tensor(true)]; + tensor attn_39_cast_fp16 = matmul(transpose_x = attn_39_transpose_x_0, transpose_y = attn_39_transpose_y_0, x = var_7595_cast_fp16, y = var_7593_cast_fp16)[name = tensor("attn_39_cast_fp16")]; + tensor var_7598 = const()[name = tensor("op_7598"), val = tensor([1, 1024, 1, 188])]; + tensor input_517_cast_fp16 = reshape(shape = var_7598, x = attn_39_cast_fp16)[name = tensor("input_517_cast_fp16")]; + tensor var_7608_pad_type_0 = const()[name = tensor("op_7608_pad_type_0"), val = tensor("valid")]; + tensor var_7608_strides_0 = const()[name = tensor("op_7608_strides_0"), val = tensor([1, 1])]; + tensor var_7608_pad_0 = const()[name = tensor("op_7608_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7608_dilations_0 = const()[name = tensor("op_7608_dilations_0"), val = tensor([1, 1])]; + tensor var_7608_groups_0 = const()[name = tensor("op_7608_groups_0"), val = tensor(1)]; + tensor layers_19_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(453513472))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454299968))), name = tensor("layers_19_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7608_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7608_dilations_0, groups = var_7608_groups_0, pad = var_7608_pad_0, pad_type = var_7608_pad_type_0, strides = var_7608_strides_0, weight = layers_19_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_517_cast_fp16)[name = tensor("op_7608_cast_fp16")]; + tensor var_7614_pad_type_0 = const()[name = tensor("op_7614_pad_type_0"), val = tensor("valid")]; + tensor var_7614_strides_0 = const()[name = tensor("op_7614_strides_0"), val = tensor([1, 1])]; + tensor var_7614_pad_0 = const()[name = tensor("op_7614_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7614_dilations_0 = const()[name = tensor("op_7614_dilations_0"), val = tensor([1, 1])]; + tensor var_7614_groups_0 = const()[name = tensor("op_7614_groups_0"), val = tensor(1)]; + tensor layers_19_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454329920))), name = tensor("layers_19_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454300160))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7614_cast_fp16 = conv(dilations = var_7614_dilations_0, groups = var_7614_groups_0, pad = var_7614_pad_0, pad_type = var_7614_pad_type_0, strides = var_7614_strides_0, weight = layers_19_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_517_cast_fp16)[name = tensor("op_7614_cast_fp16")]; + tensor obj_81_cast_fp16 = add(x = var_7608_cast_fp16, y = var_7614_cast_fp16)[name = tensor("obj_81_cast_fp16")]; + tensor inputs_195_cast_fp16 = add(x = inputs_193_cast_fp16, y = obj_81_cast_fp16)[name = tensor("inputs_195_cast_fp16")]; + tensor out_195_axes_0 = const()[name = tensor("out_195_axes_0"), val = tensor([1])]; + tensor var_7625_to_fp16 = const()[name = tensor("op_7625_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_195_cast_fp16 = layer_norm(axes = out_195_axes_0, epsilon = var_7625_to_fp16, x = inputs_195_cast_fp16)[name = tensor("out_195_cast_fp16")]; + tensor input_519_gamma_0_to_fp16 = const()[name = tensor("input_519_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454461056)))]; + tensor input_519_beta_0_to_fp16 = const()[name = tensor("input_519_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454463168)))]; + tensor input_519_epsilon_0_to_fp16 = const()[name = tensor("input_519_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_519_cast_fp16 = batch_norm(beta = input_519_beta_0_to_fp16, epsilon = input_519_epsilon_0_to_fp16, gamma = input_519_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_195_cast_fp16)[name = tensor("input_519_cast_fp16")]; + tensor var_7646_pad_type_0 = const()[name = tensor("op_7646_pad_type_0"), val = tensor("valid")]; + tensor var_7646_strides_0 = const()[name = tensor("op_7646_strides_0"), val = tensor([1, 1])]; + tensor var_7646_pad_0 = const()[name = tensor("op_7646_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7646_dilations_0 = const()[name = tensor("op_7646_dilations_0"), val = tensor([1, 1])]; + tensor var_7646_groups_0 = const()[name = tensor("op_7646_groups_0"), val = tensor(1)]; + tensor layers_19_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454465280))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(456038208))), name = tensor("layers_19_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_7646_cast_fp16 = conv(dilations = var_7646_dilations_0, groups = var_7646_groups_0, pad = var_7646_pad_0, pad_type = var_7646_pad_type_0, strides = var_7646_strides_0, weight = layers_19_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_519_cast_fp16)[name = tensor("op_7646_cast_fp16")]; + tensor var_7652_pad_type_0 = const()[name = tensor("op_7652_pad_type_0"), val = tensor("valid")]; + tensor var_7652_strides_0 = const()[name = tensor("op_7652_strides_0"), val = tensor([1, 1])]; + tensor var_7652_pad_0 = const()[name = tensor("op_7652_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7652_dilations_0 = const()[name = tensor("op_7652_dilations_0"), val = tensor([1, 1])]; + tensor var_7652_groups_0 = const()[name = tensor("op_7652_groups_0"), val = tensor(1)]; + tensor layers_19_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(456104960))), name = tensor("layers_19_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(456038400))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_7652_cast_fp16 = conv(dilations = var_7652_dilations_0, groups = var_7652_groups_0, pad = var_7652_pad_0, pad_type = var_7652_pad_type_0, strides = var_7652_strides_0, weight = layers_19_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_519_cast_fp16)[name = tensor("op_7652_cast_fp16")]; + tensor input_521_cast_fp16 = add(x = var_7646_cast_fp16, y = var_7652_cast_fp16)[name = tensor("input_521_cast_fp16")]; + tensor input_523_split_num_splits_0 = const()[name = tensor("input_523_split_num_splits_0"), val = tensor(2)]; + tensor input_523_split_axis_0 = const()[name = tensor("input_523_split_axis_0"), val = tensor(1)]; + tensor input_523_split_cast_fp16_0, tensor input_523_split_cast_fp16_1 = split(axis = input_523_split_axis_0, num_splits = input_523_split_num_splits_0, x = input_521_cast_fp16)[name = tensor("input_523_split_cast_fp16")]; + tensor input_523_split_1_sigmoid_cast_fp16 = sigmoid(x = input_523_split_cast_fp16_1)[name = tensor("input_523_split_1_sigmoid_cast_fp16")]; + tensor input_523_cast_fp16 = mul(x = input_523_split_cast_fp16_0, y = input_523_split_1_sigmoid_cast_fp16)[name = tensor("input_523_cast_fp16")]; + tensor input_525_pad_type_0 = const()[name = tensor("input_525_pad_type_0"), val = tensor("custom")]; + tensor input_525_pad_0 = const()[name = tensor("input_525_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_525_groups_0 = const()[name = tensor("input_525_groups_0"), val = tensor(1024)]; + tensor input_525_strides_0 = const()[name = tensor("input_525_strides_0"), val = tensor([1, 1])]; + tensor input_525_dilations_0 = const()[name = tensor("input_525_dilations_0"), val = tensor([1, 1])]; + tensor const_306_to_fp16 = const()[name = tensor("const_306_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(456367168)))]; + tensor const_307_to_fp16 = const()[name = tensor("const_307_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(456385664)))]; + tensor input_527_cast_fp16 = conv(bias = const_307_to_fp16, dilations = input_525_dilations_0, groups = input_525_groups_0, pad = input_525_pad_0, pad_type = input_525_pad_type_0, strides = input_525_strides_0, weight = const_306_to_fp16, x = input_523_cast_fp16)[name = tensor("input_527_cast_fp16")]; + tensor input_529_cast_fp16 = silu(x = input_527_cast_fp16)[name = tensor("input_529_cast_fp16")]; + tensor var_7674_pad_type_0 = const()[name = tensor("op_7674_pad_type_0"), val = tensor("valid")]; + tensor var_7674_strides_0 = const()[name = tensor("op_7674_strides_0"), val = tensor([1, 1])]; + tensor var_7674_pad_0 = const()[name = tensor("op_7674_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7674_dilations_0 = const()[name = tensor("op_7674_dilations_0"), val = tensor([1, 1])]; + tensor var_7674_groups_0 = const()[name = tensor("op_7674_groups_0"), val = tensor(1)]; + tensor layers_19_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(456387776))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(457174272))), name = tensor("layers_19_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7674_cast_fp16 = conv(dilations = var_7674_dilations_0, groups = var_7674_groups_0, pad = var_7674_pad_0, pad_type = var_7674_pad_type_0, strides = var_7674_strides_0, weight = layers_19_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_529_cast_fp16)[name = tensor("op_7674_cast_fp16")]; + tensor var_7680_pad_type_0 = const()[name = tensor("op_7680_pad_type_0"), val = tensor("valid")]; + tensor var_7680_strides_0 = const()[name = tensor("op_7680_strides_0"), val = tensor([1, 1])]; + tensor var_7680_pad_0 = const()[name = tensor("op_7680_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7680_dilations_0 = const()[name = tensor("op_7680_dilations_0"), val = tensor([1, 1])]; + tensor var_7680_groups_0 = const()[name = tensor("op_7680_groups_0"), val = tensor(1)]; + tensor layers_19_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(457205120))), name = tensor("layers_19_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(457174464))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7680_cast_fp16 = conv(dilations = var_7680_dilations_0, groups = var_7680_groups_0, pad = var_7680_pad_0, pad_type = var_7680_pad_type_0, strides = var_7680_strides_0, weight = layers_19_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_529_cast_fp16)[name = tensor("op_7680_cast_fp16")]; + tensor x_119_cast_fp16 = add(x = var_7674_cast_fp16, y = var_7680_cast_fp16)[name = tensor("x_119_cast_fp16")]; + tensor inputs_197_cast_fp16 = add(x = inputs_195_cast_fp16, y = x_119_cast_fp16)[name = tensor("inputs_197_cast_fp16")]; + tensor out_197_axes_0 = const()[name = tensor("out_197_axes_0"), val = tensor([1])]; + tensor var_7691_to_fp16 = const()[name = tensor("op_7691_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_197_cast_fp16 = layer_norm(axes = out_197_axes_0, epsilon = var_7691_to_fp16, x = inputs_197_cast_fp16)[name = tensor("out_197_cast_fp16")]; + tensor input_531_gamma_0_to_fp16 = const()[name = tensor("input_531_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(457336256)))]; + tensor input_531_beta_0_to_fp16 = const()[name = tensor("input_531_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(457338368)))]; + tensor input_531_epsilon_0_to_fp16 = const()[name = tensor("input_531_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_531_cast_fp16 = batch_norm(beta = input_531_beta_0_to_fp16, epsilon = input_531_epsilon_0_to_fp16, gamma = input_531_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_197_cast_fp16)[name = tensor("input_531_cast_fp16")]; + tensor var_7711_pad_type_0 = const()[name = tensor("op_7711_pad_type_0"), val = tensor("valid")]; + tensor var_7711_strides_0 = const()[name = tensor("op_7711_strides_0"), val = tensor([1, 1])]; + tensor var_7711_pad_0 = const()[name = tensor("op_7711_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7711_dilations_0 = const()[name = tensor("op_7711_dilations_0"), val = tensor([1, 1])]; + tensor var_7711_groups_0 = const()[name = tensor("op_7711_groups_0"), val = tensor(1)]; + tensor layers_19_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(457340480))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460486272))), name = tensor("layers_19_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_7711_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_7711_dilations_0, groups = var_7711_groups_0, pad = var_7711_pad_0, pad_type = var_7711_pad_type_0, strides = var_7711_strides_0, weight = layers_19_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_531_cast_fp16)[name = tensor("op_7711_cast_fp16")]; + tensor var_7717_pad_type_0 = const()[name = tensor("op_7717_pad_type_0"), val = tensor("valid")]; + tensor var_7717_strides_0 = const()[name = tensor("op_7717_strides_0"), val = tensor([1, 1])]; + tensor var_7717_pad_0 = const()[name = tensor("op_7717_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7717_dilations_0 = const()[name = tensor("op_7717_dilations_0"), val = tensor([1, 1])]; + tensor var_7717_groups_0 = const()[name = tensor("op_7717_groups_0"), val = tensor(1)]; + tensor layers_19_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460614144))), name = tensor("layers_19_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460486464))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_7717_cast_fp16 = conv(dilations = var_7717_dilations_0, groups = var_7717_groups_0, pad = var_7717_pad_0, pad_type = var_7717_pad_type_0, strides = var_7717_strides_0, weight = layers_19_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_531_cast_fp16)[name = tensor("op_7717_cast_fp16")]; + tensor input_533_cast_fp16 = add(x = var_7711_cast_fp16, y = var_7717_cast_fp16)[name = tensor("input_533_cast_fp16")]; + tensor input_535_cast_fp16 = silu(x = input_533_cast_fp16)[name = tensor("input_535_cast_fp16")]; + tensor var_7728_pad_type_0 = const()[name = tensor("op_7728_pad_type_0"), val = tensor("valid")]; + tensor var_7728_strides_0 = const()[name = tensor("op_7728_strides_0"), val = tensor([1, 1])]; + tensor var_7728_pad_0 = const()[name = tensor("op_7728_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7728_dilations_0 = const()[name = tensor("op_7728_dilations_0"), val = tensor([1, 1])]; + tensor var_7728_groups_0 = const()[name = tensor("op_7728_groups_0"), val = tensor(1)]; + tensor layers_19_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(461138496))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(464284288))), name = tensor("layers_19_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_7728_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7728_dilations_0, groups = var_7728_groups_0, pad = var_7728_pad_0, pad_type = var_7728_pad_type_0, strides = var_7728_strides_0, weight = layers_19_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_535_cast_fp16)[name = tensor("op_7728_cast_fp16")]; + tensor var_7734_pad_type_0 = const()[name = tensor("op_7734_pad_type_0"), val = tensor("valid")]; + tensor var_7734_strides_0 = const()[name = tensor("op_7734_strides_0"), val = tensor([1, 1])]; + tensor var_7734_pad_0 = const()[name = tensor("op_7734_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7734_dilations_0 = const()[name = tensor("op_7734_dilations_0"), val = tensor([1, 1])]; + tensor var_7734_groups_0 = const()[name = tensor("op_7734_groups_0"), val = tensor(1)]; + tensor layers_19_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(464423360))), name = tensor("layers_19_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(464284480))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_7734_cast_fp16 = conv(dilations = var_7734_dilations_0, groups = var_7734_groups_0, pad = var_7734_pad_0, pad_type = var_7734_pad_type_0, strides = var_7734_strides_0, weight = layers_19_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_535_cast_fp16)[name = tensor("op_7734_cast_fp16")]; + tensor x_121_cast_fp16 = add(x = var_7728_cast_fp16, y = var_7734_cast_fp16)[name = tensor("x_121_cast_fp16")]; + tensor var_7736_to_fp16 = const()[name = tensor("op_7736_to_fp16"), val = tensor(0x1p-1)]; + tensor var_7737_cast_fp16 = mul(x = x_121_cast_fp16, y = var_7736_to_fp16)[name = tensor("op_7737_cast_fp16")]; + tensor inputs_199_cast_fp16 = add(x = inputs_197_cast_fp16, y = var_7737_cast_fp16)[name = tensor("inputs_199_cast_fp16")]; + tensor out_199_axes_0 = const()[name = tensor("out_199_axes_0"), val = tensor([1])]; + tensor var_7747_to_fp16 = const()[name = tensor("op_7747_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_199_cast_fp16 = layer_norm(axes = out_199_axes_0, epsilon = var_7747_to_fp16, x = inputs_199_cast_fp16)[name = tensor("out_199_cast_fp16")]; + tensor inputs_201_gamma_0_to_fp16 = const()[name = tensor("inputs_201_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(464947712)))]; + tensor inputs_201_beta_0_to_fp16 = const()[name = tensor("inputs_201_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(464949824)))]; + tensor inputs_201_epsilon_0_to_fp16 = const()[name = tensor("inputs_201_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_201_cast_fp16 = batch_norm(beta = inputs_201_beta_0_to_fp16, epsilon = inputs_201_epsilon_0_to_fp16, gamma = inputs_201_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_199_cast_fp16)[name = tensor("inputs_201_cast_fp16")]; + tensor var_7761 = const()[name = tensor("op_7761"), val = tensor(3)]; + tensor out_201_axes_0 = const()[name = tensor("out_201_axes_0"), val = tensor([1])]; + tensor var_7792_to_fp16 = const()[name = tensor("op_7792_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_201_cast_fp16 = layer_norm(axes = out_201_axes_0, epsilon = var_7792_to_fp16, x = inputs_201_cast_fp16)[name = tensor("out_201_cast_fp16")]; + tensor input_537_gamma_0_to_fp16 = const()[name = tensor("input_537_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(464951936)))]; + tensor input_537_beta_0_to_fp16 = const()[name = tensor("input_537_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(464954048)))]; + tensor input_537_epsilon_0_to_fp16 = const()[name = tensor("input_537_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_537_cast_fp16 = batch_norm(beta = input_537_beta_0_to_fp16, epsilon = input_537_epsilon_0_to_fp16, gamma = input_537_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_201_cast_fp16)[name = tensor("input_537_cast_fp16")]; + tensor var_7812_pad_type_0 = const()[name = tensor("op_7812_pad_type_0"), val = tensor("valid")]; + tensor var_7812_strides_0 = const()[name = tensor("op_7812_strides_0"), val = tensor([1, 1])]; + tensor var_7812_pad_0 = const()[name = tensor("op_7812_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7812_dilations_0 = const()[name = tensor("op_7812_dilations_0"), val = tensor([1, 1])]; + tensor var_7812_groups_0 = const()[name = tensor("op_7812_groups_0"), val = tensor(1)]; + tensor layers_20_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(464956160))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(468101952))), name = tensor("layers_20_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_7812_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_7812_dilations_0, groups = var_7812_groups_0, pad = var_7812_pad_0, pad_type = var_7812_pad_type_0, strides = var_7812_strides_0, weight = layers_20_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_537_cast_fp16)[name = tensor("op_7812_cast_fp16")]; + tensor var_7818_pad_type_0 = const()[name = tensor("op_7818_pad_type_0"), val = tensor("valid")]; + tensor var_7818_strides_0 = const()[name = tensor("op_7818_strides_0"), val = tensor([1, 1])]; + tensor var_7818_pad_0 = const()[name = tensor("op_7818_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7818_dilations_0 = const()[name = tensor("op_7818_dilations_0"), val = tensor([1, 1])]; + tensor var_7818_groups_0 = const()[name = tensor("op_7818_groups_0"), val = tensor(1)]; + tensor layers_20_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(468232064))), name = tensor("layers_20_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(468102144))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_7818_cast_fp16 = conv(dilations = var_7818_dilations_0, groups = var_7818_groups_0, pad = var_7818_pad_0, pad_type = var_7818_pad_type_0, strides = var_7818_strides_0, weight = layers_20_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_537_cast_fp16)[name = tensor("op_7818_cast_fp16")]; + tensor input_539_cast_fp16 = add(x = var_7812_cast_fp16, y = var_7818_cast_fp16)[name = tensor("input_539_cast_fp16")]; + tensor input_541_cast_fp16 = silu(x = input_539_cast_fp16)[name = tensor("input_541_cast_fp16")]; + tensor var_7829_pad_type_0 = const()[name = tensor("op_7829_pad_type_0"), val = tensor("valid")]; + tensor var_7829_strides_0 = const()[name = tensor("op_7829_strides_0"), val = tensor([1, 1])]; + tensor var_7829_pad_0 = const()[name = tensor("op_7829_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7829_dilations_0 = const()[name = tensor("op_7829_dilations_0"), val = tensor([1, 1])]; + tensor var_7829_groups_0 = const()[name = tensor("op_7829_groups_0"), val = tensor(1)]; + tensor layers_20_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(468756416))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471902208))), name = tensor("layers_20_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_7829_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7829_dilations_0, groups = var_7829_groups_0, pad = var_7829_pad_0, pad_type = var_7829_pad_type_0, strides = var_7829_strides_0, weight = layers_20_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_541_cast_fp16)[name = tensor("op_7829_cast_fp16")]; + tensor var_7835_pad_type_0 = const()[name = tensor("op_7835_pad_type_0"), val = tensor("valid")]; + tensor var_7835_strides_0 = const()[name = tensor("op_7835_strides_0"), val = tensor([1, 1])]; + tensor var_7835_pad_0 = const()[name = tensor("op_7835_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7835_dilations_0 = const()[name = tensor("op_7835_dilations_0"), val = tensor([1, 1])]; + tensor var_7835_groups_0 = const()[name = tensor("op_7835_groups_0"), val = tensor(1)]; + tensor layers_20_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(472049856))), name = tensor("layers_20_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471902400))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_7835_cast_fp16 = conv(dilations = var_7835_dilations_0, groups = var_7835_groups_0, pad = var_7835_pad_0, pad_type = var_7835_pad_type_0, strides = var_7835_strides_0, weight = layers_20_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_541_cast_fp16)[name = tensor("op_7835_cast_fp16")]; + tensor x_123_cast_fp16 = add(x = var_7829_cast_fp16, y = var_7835_cast_fp16)[name = tensor("x_123_cast_fp16")]; + tensor var_7837_to_fp16 = const()[name = tensor("op_7837_to_fp16"), val = tensor(0x1p-1)]; + tensor var_7838_cast_fp16 = mul(x = x_123_cast_fp16, y = var_7837_to_fp16)[name = tensor("op_7838_cast_fp16")]; + tensor inputs_203_cast_fp16 = add(x = inputs_201_cast_fp16, y = var_7838_cast_fp16)[name = tensor("inputs_203_cast_fp16")]; + tensor out_203_axes_0 = const()[name = tensor("out_203_axes_0"), val = tensor([1])]; + tensor var_7848_to_fp16 = const()[name = tensor("op_7848_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_203_cast_fp16 = layer_norm(axes = out_203_axes_0, epsilon = var_7848_to_fp16, x = inputs_203_cast_fp16)[name = tensor("out_203_cast_fp16")]; + tensor obj_83_gamma_0_to_fp16 = const()[name = tensor("obj_83_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(472574208)))]; + tensor obj_83_beta_0_to_fp16 = const()[name = tensor("obj_83_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(472576320)))]; + tensor obj_83_epsilon_0_to_fp16 = const()[name = tensor("obj_83_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_83_cast_fp16 = batch_norm(beta = obj_83_beta_0_to_fp16, epsilon = obj_83_epsilon_0_to_fp16, gamma = obj_83_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_203_cast_fp16)[name = tensor("obj_83_cast_fp16")]; + tensor var_7873_pad_type_0 = const()[name = tensor("op_7873_pad_type_0"), val = tensor("valid")]; + tensor var_7873_strides_0 = const()[name = tensor("op_7873_strides_0"), val = tensor([1, 1])]; + tensor var_7873_pad_0 = const()[name = tensor("op_7873_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7873_dilations_0 = const()[name = tensor("op_7873_dilations_0"), val = tensor([1, 1])]; + tensor var_7873_groups_0 = const()[name = tensor("op_7873_groups_0"), val = tensor(1)]; + tensor layers_20_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(472578432))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473364928))), name = tensor("layers_20_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7873_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7873_dilations_0, groups = var_7873_groups_0, pad = var_7873_pad_0, pad_type = var_7873_pad_type_0, strides = var_7873_strides_0, weight = layers_20_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_83_cast_fp16)[name = tensor("op_7873_cast_fp16")]; + tensor var_7879_pad_type_0 = const()[name = tensor("op_7879_pad_type_0"), val = tensor("valid")]; + tensor var_7879_strides_0 = const()[name = tensor("op_7879_strides_0"), val = tensor([1, 1])]; + tensor var_7879_pad_0 = const()[name = tensor("op_7879_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7879_dilations_0 = const()[name = tensor("op_7879_dilations_0"), val = tensor([1, 1])]; + tensor var_7879_groups_0 = const()[name = tensor("op_7879_groups_0"), val = tensor(1)]; + tensor layers_20_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473395392))), name = tensor("layers_20_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473365120))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7879_cast_fp16 = conv(dilations = var_7879_dilations_0, groups = var_7879_groups_0, pad = var_7879_pad_0, pad_type = var_7879_pad_type_0, strides = var_7879_strides_0, weight = layers_20_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_83_cast_fp16)[name = tensor("op_7879_cast_fp16")]; + tensor query_81_cast_fp16 = add(x = var_7873_cast_fp16, y = var_7879_cast_fp16)[name = tensor("query_81_cast_fp16")]; + tensor var_7888_pad_type_0 = const()[name = tensor("op_7888_pad_type_0"), val = tensor("valid")]; + tensor var_7888_strides_0 = const()[name = tensor("op_7888_strides_0"), val = tensor([1, 1])]; + tensor var_7888_pad_0 = const()[name = tensor("op_7888_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7888_dilations_0 = const()[name = tensor("op_7888_dilations_0"), val = tensor([1, 1])]; + tensor var_7888_groups_0 = const()[name = tensor("op_7888_groups_0"), val = tensor(1)]; + tensor layers_20_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473526528))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(474313024))), name = tensor("layers_20_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7888_cast_fp16 = conv(dilations = var_7888_dilations_0, groups = var_7888_groups_0, pad = var_7888_pad_0, pad_type = var_7888_pad_type_0, strides = var_7888_strides_0, weight = layers_20_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_83_cast_fp16)[name = tensor("op_7888_cast_fp16")]; + tensor var_7894_pad_type_0 = const()[name = tensor("op_7894_pad_type_0"), val = tensor("valid")]; + tensor var_7894_strides_0 = const()[name = tensor("op_7894_strides_0"), val = tensor([1, 1])]; + tensor var_7894_pad_0 = const()[name = tensor("op_7894_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7894_dilations_0 = const()[name = tensor("op_7894_dilations_0"), val = tensor([1, 1])]; + tensor var_7894_groups_0 = const()[name = tensor("op_7894_groups_0"), val = tensor(1)]; + tensor layers_20_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(474344896))), name = tensor("layers_20_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(474313216))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7894_cast_fp16 = conv(dilations = var_7894_dilations_0, groups = var_7894_groups_0, pad = var_7894_pad_0, pad_type = var_7894_pad_type_0, strides = var_7894_strides_0, weight = layers_20_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_83_cast_fp16)[name = tensor("op_7894_cast_fp16")]; + tensor key_41_cast_fp16 = add(x = var_7888_cast_fp16, y = var_7894_cast_fp16)[name = tensor("key_41_cast_fp16")]; + tensor var_7904_pad_type_0 = const()[name = tensor("op_7904_pad_type_0"), val = tensor("valid")]; + tensor var_7904_strides_0 = const()[name = tensor("op_7904_strides_0"), val = tensor([1, 1])]; + tensor var_7904_pad_0 = const()[name = tensor("op_7904_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7904_dilations_0 = const()[name = tensor("op_7904_dilations_0"), val = tensor([1, 1])]; + tensor var_7904_groups_0 = const()[name = tensor("op_7904_groups_0"), val = tensor(1)]; + tensor layers_20_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(474476032))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(475262528))), name = tensor("layers_20_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7904_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7904_dilations_0, groups = var_7904_groups_0, pad = var_7904_pad_0, pad_type = var_7904_pad_type_0, strides = var_7904_strides_0, weight = layers_20_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_83_cast_fp16)[name = tensor("op_7904_cast_fp16")]; + tensor var_7910_pad_type_0 = const()[name = tensor("op_7910_pad_type_0"), val = tensor("valid")]; + tensor var_7910_strides_0 = const()[name = tensor("op_7910_strides_0"), val = tensor([1, 1])]; + tensor var_7910_pad_0 = const()[name = tensor("op_7910_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7910_dilations_0 = const()[name = tensor("op_7910_dilations_0"), val = tensor([1, 1])]; + tensor var_7910_groups_0 = const()[name = tensor("op_7910_groups_0"), val = tensor(1)]; + tensor layers_20_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(475291392))), name = tensor("layers_20_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(475262720))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7910_cast_fp16 = conv(dilations = var_7910_dilations_0, groups = var_7910_groups_0, pad = var_7910_pad_0, pad_type = var_7910_pad_type_0, strides = var_7910_strides_0, weight = layers_20_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_83_cast_fp16)[name = tensor("op_7910_cast_fp16")]; + tensor value_41_cast_fp16 = add(x = var_7904_cast_fp16, y = var_7910_cast_fp16)[name = tensor("value_41_cast_fp16")]; + tensor var_7913_to_fp16 = const()[name = tensor("op_7913_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(475422528)))]; + tensor query_83_cast_fp16 = add(x = query_81_cast_fp16, y = var_7913_to_fp16)[name = tensor("query_83_cast_fp16")]; + tensor var_7916_to_fp16 = const()[name = tensor("op_7916_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(475424640)))]; + tensor q_with_bias_v_41_cast_fp16 = add(x = query_81_cast_fp16, y = var_7916_to_fp16)[name = tensor("q_with_bias_v_41_cast_fp16")]; + tensor var_7926_pad_type_0 = const()[name = tensor("op_7926_pad_type_0"), val = tensor("valid")]; + tensor var_7926_strides_0 = const()[name = tensor("op_7926_strides_0"), val = tensor([1, 1])]; + tensor var_7926_pad_0 = const()[name = tensor("op_7926_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7926_dilations_0 = const()[name = tensor("op_7926_dilations_0"), val = tensor([1, 1])]; + tensor var_7926_groups_0 = const()[name = tensor("op_7926_groups_0"), val = tensor(1)]; + tensor layers_20_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(475426752))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(476213248))), name = tensor("layers_20_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7926_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7926_dilations_0, groups = var_7926_groups_0, pad = var_7926_pad_0, pad_type = var_7926_pad_type_0, strides = var_7926_strides_0, weight = layers_20_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_7926_cast_fp16")]; + tensor var_7932_pad_type_0 = const()[name = tensor("op_7932_pad_type_0"), val = tensor("valid")]; + tensor var_7932_strides_0 = const()[name = tensor("op_7932_strides_0"), val = tensor([1, 1])]; + tensor var_7932_pad_0 = const()[name = tensor("op_7932_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7932_dilations_0 = const()[name = tensor("op_7932_dilations_0"), val = tensor([1, 1])]; + tensor var_7932_groups_0 = const()[name = tensor("op_7932_groups_0"), val = tensor(1)]; + tensor layers_20_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(476283776))), name = tensor("layers_20_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(476213440))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7932_cast_fp16 = conv(dilations = var_7932_dilations_0, groups = var_7932_groups_0, pad = var_7932_pad_0, pad_type = var_7932_pad_type_0, strides = var_7932_strides_0, weight = layers_20_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_7932_cast_fp16")]; + tensor p_41_cast_fp16 = add(x = var_7926_cast_fp16, y = var_7932_cast_fp16)[name = tensor("p_41_cast_fp16")]; + tensor var_7936 = const()[name = tensor("op_7936"), val = tensor([1, 8, 128, 188])]; + tensor var_7937_cast_fp16 = reshape(shape = var_7936, x = q_with_bias_v_41_cast_fp16)[name = tensor("op_7937_cast_fp16")]; + tensor var_7938 = const()[name = tensor("op_7938"), val = tensor([1, 8, 128, -1])]; + tensor var_7939_cast_fp16 = reshape(shape = var_7938, x = p_41_cast_fp16)[name = tensor("op_7939_cast_fp16")]; + tensor matrix_bd_161_transpose_x_0 = const()[name = tensor("matrix_bd_161_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_161_transpose_y_0 = const()[name = tensor("matrix_bd_161_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_161_cast_fp16 = matmul(transpose_x = matrix_bd_161_transpose_x_0, transpose_y = matrix_bd_161_transpose_y_0, x = var_7937_cast_fp16, y = var_7939_cast_fp16)[name = tensor("matrix_bd_161_cast_fp16")]; + tensor matrix_bd_163_pad_0 = const()[name = tensor("matrix_bd_163_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_163_mode_0 = const()[name = tensor("matrix_bd_163_mode_0"), val = tensor("constant")]; + tensor const_230_to_fp16 = const()[name = tensor("const_230_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_163_cast_fp16 = pad(constant_val = const_230_to_fp16, mode = matrix_bd_163_mode_0, pad = matrix_bd_163_pad_0, x = matrix_bd_161_cast_fp16)[name = tensor("matrix_bd_163_cast_fp16")]; + tensor var_7948 = const()[name = tensor("op_7948"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_165_cast_fp16 = reshape(shape = var_7948, x = matrix_bd_163_cast_fp16)[name = tensor("matrix_bd_165_cast_fp16")]; + tensor var_7952_begin_0 = const()[name = tensor("op_7952_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_7952_end_0 = const()[name = tensor("op_7952_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_7952_end_mask_0 = const()[name = tensor("op_7952_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_7952_cast_fp16 = slice_by_index(begin = var_7952_begin_0, end = var_7952_end_0, end_mask = var_7952_end_mask_0, x = matrix_bd_165_cast_fp16)[name = tensor("op_7952_cast_fp16")]; + tensor var_7953 = const()[name = tensor("op_7953"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_167_cast_fp16 = reshape(shape = var_7953, x = var_7952_cast_fp16)[name = tensor("matrix_bd_167_cast_fp16")]; + tensor var_7958_begin_0 = const()[name = tensor("op_7958_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7958_end_0 = const()[name = tensor("op_7958_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_7958_end_mask_0 = const()[name = tensor("op_7958_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_7958_cast_fp16 = slice_by_index(begin = var_7958_begin_0, end = var_7958_end_0, end_mask = var_7958_end_mask_0, x = matrix_bd_167_cast_fp16)[name = tensor("op_7958_cast_fp16")]; + tensor var_7959_to_fp16 = const()[name = tensor("op_7959_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_41_cast_fp16 = mul(x = var_7958_cast_fp16, y = var_7959_to_fp16)[name = tensor("qk_mask_41_cast_fp16")]; + tensor var_7963 = const()[name = tensor("op_7963"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_41_cast_fp16 = reshape(shape = var_7963, x = query_83_cast_fp16)[name = tensor("mh_q_41_cast_fp16")]; + tensor var_7965_to_fp16 = const()[name = tensor("op_7965_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_7966_cast_fp16 = mul(x = mh_q_41_cast_fp16, y = var_7965_to_fp16)[name = tensor("op_7966_cast_fp16")]; + tensor var_7969 = const()[name = tensor("op_7969"), val = tensor([1, 8, 128, 188])]; + tensor var_7970_cast_fp16 = reshape(shape = var_7969, x = key_41_cast_fp16)[name = tensor("op_7970_cast_fp16")]; + tensor mh_w_81_transpose_x_0 = const()[name = tensor("mh_w_81_transpose_x_0"), val = tensor(true)]; + tensor mh_w_81_transpose_y_0 = const()[name = tensor("mh_w_81_transpose_y_0"), val = tensor(false)]; + tensor mh_w_81_cast_fp16 = matmul(transpose_x = mh_w_81_transpose_x_0, transpose_y = mh_w_81_transpose_y_0, x = var_7966_cast_fp16, y = var_7970_cast_fp16)[name = tensor("mh_w_81_cast_fp16")]; + tensor mh_w_83_cast_fp16 = add(x = mh_w_81_cast_fp16, y = qk_mask_41_cast_fp16)[name = tensor("mh_w_83_cast_fp16")]; + tensor var_7974_cast_fp16 = softmax(axis = var_7761, x = mh_w_83_cast_fp16)[name = tensor("op_7974_cast_fp16")]; + tensor var_7975 = const()[name = tensor("op_7975"), val = tensor([1, 8, 128, 188])]; + tensor var_7976_cast_fp16 = reshape(shape = var_7975, x = value_41_cast_fp16)[name = tensor("op_7976_cast_fp16")]; + tensor attn_41_transpose_x_0 = const()[name = tensor("attn_41_transpose_x_0"), val = tensor(false)]; + tensor attn_41_transpose_y_0 = const()[name = tensor("attn_41_transpose_y_0"), val = tensor(true)]; + tensor attn_41_cast_fp16 = matmul(transpose_x = attn_41_transpose_x_0, transpose_y = attn_41_transpose_y_0, x = var_7976_cast_fp16, y = var_7974_cast_fp16)[name = tensor("attn_41_cast_fp16")]; + tensor var_7979 = const()[name = tensor("op_7979"), val = tensor([1, 1024, 1, 188])]; + tensor input_543_cast_fp16 = reshape(shape = var_7979, x = attn_41_cast_fp16)[name = tensor("input_543_cast_fp16")]; + tensor var_7989_pad_type_0 = const()[name = tensor("op_7989_pad_type_0"), val = tensor("valid")]; + tensor var_7989_strides_0 = const()[name = tensor("op_7989_strides_0"), val = tensor([1, 1])]; + tensor var_7989_pad_0 = const()[name = tensor("op_7989_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7989_dilations_0 = const()[name = tensor("op_7989_dilations_0"), val = tensor([1, 1])]; + tensor var_7989_groups_0 = const()[name = tensor("op_7989_groups_0"), val = tensor(1)]; + tensor layers_20_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(476414912))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(477201408))), name = tensor("layers_20_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7989_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_7989_dilations_0, groups = var_7989_groups_0, pad = var_7989_pad_0, pad_type = var_7989_pad_type_0, strides = var_7989_strides_0, weight = layers_20_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_543_cast_fp16)[name = tensor("op_7989_cast_fp16")]; + tensor var_7995_pad_type_0 = const()[name = tensor("op_7995_pad_type_0"), val = tensor("valid")]; + tensor var_7995_strides_0 = const()[name = tensor("op_7995_strides_0"), val = tensor([1, 1])]; + tensor var_7995_pad_0 = const()[name = tensor("op_7995_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_7995_dilations_0 = const()[name = tensor("op_7995_dilations_0"), val = tensor([1, 1])]; + tensor var_7995_groups_0 = const()[name = tensor("op_7995_groups_0"), val = tensor(1)]; + tensor layers_20_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(477231040))), name = tensor("layers_20_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(477201600))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_7995_cast_fp16 = conv(dilations = var_7995_dilations_0, groups = var_7995_groups_0, pad = var_7995_pad_0, pad_type = var_7995_pad_type_0, strides = var_7995_strides_0, weight = layers_20_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_543_cast_fp16)[name = tensor("op_7995_cast_fp16")]; + tensor obj_85_cast_fp16 = add(x = var_7989_cast_fp16, y = var_7995_cast_fp16)[name = tensor("obj_85_cast_fp16")]; + tensor inputs_205_cast_fp16 = add(x = inputs_203_cast_fp16, y = obj_85_cast_fp16)[name = tensor("inputs_205_cast_fp16")]; + tensor out_205_axes_0 = const()[name = tensor("out_205_axes_0"), val = tensor([1])]; + tensor var_8006_to_fp16 = const()[name = tensor("op_8006_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_205_cast_fp16 = layer_norm(axes = out_205_axes_0, epsilon = var_8006_to_fp16, x = inputs_205_cast_fp16)[name = tensor("out_205_cast_fp16")]; + tensor input_545_gamma_0_to_fp16 = const()[name = tensor("input_545_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(477362176)))]; + tensor input_545_beta_0_to_fp16 = const()[name = tensor("input_545_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(477364288)))]; + tensor input_545_epsilon_0_to_fp16 = const()[name = tensor("input_545_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_545_cast_fp16 = batch_norm(beta = input_545_beta_0_to_fp16, epsilon = input_545_epsilon_0_to_fp16, gamma = input_545_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_205_cast_fp16)[name = tensor("input_545_cast_fp16")]; + tensor var_8027_pad_type_0 = const()[name = tensor("op_8027_pad_type_0"), val = tensor("valid")]; + tensor var_8027_strides_0 = const()[name = tensor("op_8027_strides_0"), val = tensor([1, 1])]; + tensor var_8027_pad_0 = const()[name = tensor("op_8027_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8027_dilations_0 = const()[name = tensor("op_8027_dilations_0"), val = tensor([1, 1])]; + tensor var_8027_groups_0 = const()[name = tensor("op_8027_groups_0"), val = tensor(1)]; + tensor layers_20_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(477366400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(478939328))), name = tensor("layers_20_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_8027_cast_fp16 = conv(dilations = var_8027_dilations_0, groups = var_8027_groups_0, pad = var_8027_pad_0, pad_type = var_8027_pad_type_0, strides = var_8027_strides_0, weight = layers_20_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_545_cast_fp16)[name = tensor("op_8027_cast_fp16")]; + tensor var_8033_pad_type_0 = const()[name = tensor("op_8033_pad_type_0"), val = tensor("valid")]; + tensor var_8033_strides_0 = const()[name = tensor("op_8033_strides_0"), val = tensor([1, 1])]; + tensor var_8033_pad_0 = const()[name = tensor("op_8033_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8033_dilations_0 = const()[name = tensor("op_8033_dilations_0"), val = tensor([1, 1])]; + tensor var_8033_groups_0 = const()[name = tensor("op_8033_groups_0"), val = tensor(1)]; + tensor layers_20_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(479004544))), name = tensor("layers_20_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(478939520))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_8033_cast_fp16 = conv(dilations = var_8033_dilations_0, groups = var_8033_groups_0, pad = var_8033_pad_0, pad_type = var_8033_pad_type_0, strides = var_8033_strides_0, weight = layers_20_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_545_cast_fp16)[name = tensor("op_8033_cast_fp16")]; + tensor input_547_cast_fp16 = add(x = var_8027_cast_fp16, y = var_8033_cast_fp16)[name = tensor("input_547_cast_fp16")]; + tensor input_549_split_num_splits_0 = const()[name = tensor("input_549_split_num_splits_0"), val = tensor(2)]; + tensor input_549_split_axis_0 = const()[name = tensor("input_549_split_axis_0"), val = tensor(1)]; + tensor input_549_split_cast_fp16_0, tensor input_549_split_cast_fp16_1 = split(axis = input_549_split_axis_0, num_splits = input_549_split_num_splits_0, x = input_547_cast_fp16)[name = tensor("input_549_split_cast_fp16")]; + tensor input_549_split_1_sigmoid_cast_fp16 = sigmoid(x = input_549_split_cast_fp16_1)[name = tensor("input_549_split_1_sigmoid_cast_fp16")]; + tensor input_549_cast_fp16 = mul(x = input_549_split_cast_fp16_0, y = input_549_split_1_sigmoid_cast_fp16)[name = tensor("input_549_cast_fp16")]; + tensor input_551_pad_type_0 = const()[name = tensor("input_551_pad_type_0"), val = tensor("custom")]; + tensor input_551_pad_0 = const()[name = tensor("input_551_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_551_groups_0 = const()[name = tensor("input_551_groups_0"), val = tensor(1024)]; + tensor input_551_strides_0 = const()[name = tensor("input_551_strides_0"), val = tensor([1, 1])]; + tensor input_551_dilations_0 = const()[name = tensor("input_551_dilations_0"), val = tensor([1, 1])]; + tensor const_308_to_fp16 = const()[name = tensor("const_308_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(479266752)))]; + tensor const_309_to_fp16 = const()[name = tensor("const_309_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(479285248)))]; + tensor input_553_cast_fp16 = conv(bias = const_309_to_fp16, dilations = input_551_dilations_0, groups = input_551_groups_0, pad = input_551_pad_0, pad_type = input_551_pad_type_0, strides = input_551_strides_0, weight = const_308_to_fp16, x = input_549_cast_fp16)[name = tensor("input_553_cast_fp16")]; + tensor input_555_cast_fp16 = silu(x = input_553_cast_fp16)[name = tensor("input_555_cast_fp16")]; + tensor var_8055_pad_type_0 = const()[name = tensor("op_8055_pad_type_0"), val = tensor("valid")]; + tensor var_8055_strides_0 = const()[name = tensor("op_8055_strides_0"), val = tensor([1, 1])]; + tensor var_8055_pad_0 = const()[name = tensor("op_8055_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8055_dilations_0 = const()[name = tensor("op_8055_dilations_0"), val = tensor([1, 1])]; + tensor var_8055_groups_0 = const()[name = tensor("op_8055_groups_0"), val = tensor(1)]; + tensor layers_20_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(479287360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(480073856))), name = tensor("layers_20_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8055_cast_fp16 = conv(dilations = var_8055_dilations_0, groups = var_8055_groups_0, pad = var_8055_pad_0, pad_type = var_8055_pad_type_0, strides = var_8055_strides_0, weight = layers_20_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_555_cast_fp16)[name = tensor("op_8055_cast_fp16")]; + tensor var_8061_pad_type_0 = const()[name = tensor("op_8061_pad_type_0"), val = tensor("valid")]; + tensor var_8061_strides_0 = const()[name = tensor("op_8061_strides_0"), val = tensor([1, 1])]; + tensor var_8061_pad_0 = const()[name = tensor("op_8061_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8061_dilations_0 = const()[name = tensor("op_8061_dilations_0"), val = tensor([1, 1])]; + tensor var_8061_groups_0 = const()[name = tensor("op_8061_groups_0"), val = tensor(1)]; + tensor layers_20_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(480105984))), name = tensor("layers_20_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(480074048))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8061_cast_fp16 = conv(dilations = var_8061_dilations_0, groups = var_8061_groups_0, pad = var_8061_pad_0, pad_type = var_8061_pad_type_0, strides = var_8061_strides_0, weight = layers_20_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_555_cast_fp16)[name = tensor("op_8061_cast_fp16")]; + tensor x_125_cast_fp16 = add(x = var_8055_cast_fp16, y = var_8061_cast_fp16)[name = tensor("x_125_cast_fp16")]; + tensor inputs_207_cast_fp16 = add(x = inputs_205_cast_fp16, y = x_125_cast_fp16)[name = tensor("inputs_207_cast_fp16")]; + tensor out_207_axes_0 = const()[name = tensor("out_207_axes_0"), val = tensor([1])]; + tensor var_8072_to_fp16 = const()[name = tensor("op_8072_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_207_cast_fp16 = layer_norm(axes = out_207_axes_0, epsilon = var_8072_to_fp16, x = inputs_207_cast_fp16)[name = tensor("out_207_cast_fp16")]; + tensor input_557_gamma_0_to_fp16 = const()[name = tensor("input_557_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(480237120)))]; + tensor input_557_beta_0_to_fp16 = const()[name = tensor("input_557_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(480239232)))]; + tensor input_557_epsilon_0_to_fp16 = const()[name = tensor("input_557_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_557_cast_fp16 = batch_norm(beta = input_557_beta_0_to_fp16, epsilon = input_557_epsilon_0_to_fp16, gamma = input_557_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_207_cast_fp16)[name = tensor("input_557_cast_fp16")]; + tensor var_8092_pad_type_0 = const()[name = tensor("op_8092_pad_type_0"), val = tensor("valid")]; + tensor var_8092_strides_0 = const()[name = tensor("op_8092_strides_0"), val = tensor([1, 1])]; + tensor var_8092_pad_0 = const()[name = tensor("op_8092_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8092_dilations_0 = const()[name = tensor("op_8092_dilations_0"), val = tensor([1, 1])]; + tensor var_8092_groups_0 = const()[name = tensor("op_8092_groups_0"), val = tensor(1)]; + tensor layers_20_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(480241344))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(483387136))), name = tensor("layers_20_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_8092_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_8092_dilations_0, groups = var_8092_groups_0, pad = var_8092_pad_0, pad_type = var_8092_pad_type_0, strides = var_8092_strides_0, weight = layers_20_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_557_cast_fp16)[name = tensor("op_8092_cast_fp16")]; + tensor var_8098_pad_type_0 = const()[name = tensor("op_8098_pad_type_0"), val = tensor("valid")]; + tensor var_8098_strides_0 = const()[name = tensor("op_8098_strides_0"), val = tensor([1, 1])]; + tensor var_8098_pad_0 = const()[name = tensor("op_8098_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8098_dilations_0 = const()[name = tensor("op_8098_dilations_0"), val = tensor([1, 1])]; + tensor var_8098_groups_0 = const()[name = tensor("op_8098_groups_0"), val = tensor(1)]; + tensor layers_20_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(483513408))), name = tensor("layers_20_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(483387328))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_8098_cast_fp16 = conv(dilations = var_8098_dilations_0, groups = var_8098_groups_0, pad = var_8098_pad_0, pad_type = var_8098_pad_type_0, strides = var_8098_strides_0, weight = layers_20_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_557_cast_fp16)[name = tensor("op_8098_cast_fp16")]; + tensor input_559_cast_fp16 = add(x = var_8092_cast_fp16, y = var_8098_cast_fp16)[name = tensor("input_559_cast_fp16")]; + tensor input_561_cast_fp16 = silu(x = input_559_cast_fp16)[name = tensor("input_561_cast_fp16")]; + tensor var_8109_pad_type_0 = const()[name = tensor("op_8109_pad_type_0"), val = tensor("valid")]; + tensor var_8109_strides_0 = const()[name = tensor("op_8109_strides_0"), val = tensor([1, 1])]; + tensor var_8109_pad_0 = const()[name = tensor("op_8109_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8109_dilations_0 = const()[name = tensor("op_8109_dilations_0"), val = tensor([1, 1])]; + tensor var_8109_groups_0 = const()[name = tensor("op_8109_groups_0"), val = tensor(1)]; + tensor layers_20_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(484037760))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(487183552))), name = tensor("layers_20_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_8109_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8109_dilations_0, groups = var_8109_groups_0, pad = var_8109_pad_0, pad_type = var_8109_pad_type_0, strides = var_8109_strides_0, weight = layers_20_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_561_cast_fp16)[name = tensor("op_8109_cast_fp16")]; + tensor var_8115_pad_type_0 = const()[name = tensor("op_8115_pad_type_0"), val = tensor("valid")]; + tensor var_8115_strides_0 = const()[name = tensor("op_8115_strides_0"), val = tensor([1, 1])]; + tensor var_8115_pad_0 = const()[name = tensor("op_8115_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8115_dilations_0 = const()[name = tensor("op_8115_dilations_0"), val = tensor([1, 1])]; + tensor var_8115_groups_0 = const()[name = tensor("op_8115_groups_0"), val = tensor(1)]; + tensor layers_20_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(487328768))), name = tensor("layers_20_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(487183744))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_8115_cast_fp16 = conv(dilations = var_8115_dilations_0, groups = var_8115_groups_0, pad = var_8115_pad_0, pad_type = var_8115_pad_type_0, strides = var_8115_strides_0, weight = layers_20_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_561_cast_fp16)[name = tensor("op_8115_cast_fp16")]; + tensor x_127_cast_fp16 = add(x = var_8109_cast_fp16, y = var_8115_cast_fp16)[name = tensor("x_127_cast_fp16")]; + tensor var_8117_to_fp16 = const()[name = tensor("op_8117_to_fp16"), val = tensor(0x1p-1)]; + tensor var_8118_cast_fp16 = mul(x = x_127_cast_fp16, y = var_8117_to_fp16)[name = tensor("op_8118_cast_fp16")]; + tensor inputs_209_cast_fp16 = add(x = inputs_207_cast_fp16, y = var_8118_cast_fp16)[name = tensor("inputs_209_cast_fp16")]; + tensor out_209_axes_0 = const()[name = tensor("out_209_axes_0"), val = tensor([1])]; + tensor var_8128_to_fp16 = const()[name = tensor("op_8128_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_209_cast_fp16 = layer_norm(axes = out_209_axes_0, epsilon = var_8128_to_fp16, x = inputs_209_cast_fp16)[name = tensor("out_209_cast_fp16")]; + tensor inputs_211_gamma_0_to_fp16 = const()[name = tensor("inputs_211_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(487853120)))]; + tensor inputs_211_beta_0_to_fp16 = const()[name = tensor("inputs_211_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(487855232)))]; + tensor inputs_211_epsilon_0_to_fp16 = const()[name = tensor("inputs_211_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_211_cast_fp16 = batch_norm(beta = inputs_211_beta_0_to_fp16, epsilon = inputs_211_epsilon_0_to_fp16, gamma = inputs_211_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_209_cast_fp16)[name = tensor("inputs_211_cast_fp16")]; + tensor var_8142 = const()[name = tensor("op_8142"), val = tensor(3)]; + tensor out_211_axes_0 = const()[name = tensor("out_211_axes_0"), val = tensor([1])]; + tensor var_8173_to_fp16 = const()[name = tensor("op_8173_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_211_cast_fp16 = layer_norm(axes = out_211_axes_0, epsilon = var_8173_to_fp16, x = inputs_211_cast_fp16)[name = tensor("out_211_cast_fp16")]; + tensor input_563_gamma_0_to_fp16 = const()[name = tensor("input_563_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(487857344)))]; + tensor input_563_beta_0_to_fp16 = const()[name = tensor("input_563_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(487859456)))]; + tensor input_563_epsilon_0_to_fp16 = const()[name = tensor("input_563_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_563_cast_fp16 = batch_norm(beta = input_563_beta_0_to_fp16, epsilon = input_563_epsilon_0_to_fp16, gamma = input_563_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_211_cast_fp16)[name = tensor("input_563_cast_fp16")]; + tensor var_8193_pad_type_0 = const()[name = tensor("op_8193_pad_type_0"), val = tensor("valid")]; + tensor var_8193_strides_0 = const()[name = tensor("op_8193_strides_0"), val = tensor([1, 1])]; + tensor var_8193_pad_0 = const()[name = tensor("op_8193_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8193_dilations_0 = const()[name = tensor("op_8193_dilations_0"), val = tensor([1, 1])]; + tensor var_8193_groups_0 = const()[name = tensor("op_8193_groups_0"), val = tensor(1)]; + tensor layers_21_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(487861568))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(491007360))), name = tensor("layers_21_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_8193_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_8193_dilations_0, groups = var_8193_groups_0, pad = var_8193_pad_0, pad_type = var_8193_pad_type_0, strides = var_8193_strides_0, weight = layers_21_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_563_cast_fp16)[name = tensor("op_8193_cast_fp16")]; + tensor var_8199_pad_type_0 = const()[name = tensor("op_8199_pad_type_0"), val = tensor("valid")]; + tensor var_8199_strides_0 = const()[name = tensor("op_8199_strides_0"), val = tensor([1, 1])]; + tensor var_8199_pad_0 = const()[name = tensor("op_8199_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8199_dilations_0 = const()[name = tensor("op_8199_dilations_0"), val = tensor([1, 1])]; + tensor var_8199_groups_0 = const()[name = tensor("op_8199_groups_0"), val = tensor(1)]; + tensor layers_21_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(491138048))), name = tensor("layers_21_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(491007552))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_8199_cast_fp16 = conv(dilations = var_8199_dilations_0, groups = var_8199_groups_0, pad = var_8199_pad_0, pad_type = var_8199_pad_type_0, strides = var_8199_strides_0, weight = layers_21_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_563_cast_fp16)[name = tensor("op_8199_cast_fp16")]; + tensor input_565_cast_fp16 = add(x = var_8193_cast_fp16, y = var_8199_cast_fp16)[name = tensor("input_565_cast_fp16")]; + tensor input_567_cast_fp16 = silu(x = input_565_cast_fp16)[name = tensor("input_567_cast_fp16")]; + tensor var_8210_pad_type_0 = const()[name = tensor("op_8210_pad_type_0"), val = tensor("valid")]; + tensor var_8210_strides_0 = const()[name = tensor("op_8210_strides_0"), val = tensor([1, 1])]; + tensor var_8210_pad_0 = const()[name = tensor("op_8210_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8210_dilations_0 = const()[name = tensor("op_8210_dilations_0"), val = tensor([1, 1])]; + tensor var_8210_groups_0 = const()[name = tensor("op_8210_groups_0"), val = tensor(1)]; + tensor layers_21_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(491662400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(494808192))), name = tensor("layers_21_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_8210_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8210_dilations_0, groups = var_8210_groups_0, pad = var_8210_pad_0, pad_type = var_8210_pad_type_0, strides = var_8210_strides_0, weight = layers_21_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_567_cast_fp16)[name = tensor("op_8210_cast_fp16")]; + tensor var_8216_pad_type_0 = const()[name = tensor("op_8216_pad_type_0"), val = tensor("valid")]; + tensor var_8216_strides_0 = const()[name = tensor("op_8216_strides_0"), val = tensor([1, 1])]; + tensor var_8216_pad_0 = const()[name = tensor("op_8216_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8216_dilations_0 = const()[name = tensor("op_8216_dilations_0"), val = tensor([1, 1])]; + tensor var_8216_groups_0 = const()[name = tensor("op_8216_groups_0"), val = tensor(1)]; + tensor layers_21_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(494973568))), name = tensor("layers_21_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(494808384))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_8216_cast_fp16 = conv(dilations = var_8216_dilations_0, groups = var_8216_groups_0, pad = var_8216_pad_0, pad_type = var_8216_pad_type_0, strides = var_8216_strides_0, weight = layers_21_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_567_cast_fp16)[name = tensor("op_8216_cast_fp16")]; + tensor x_129_cast_fp16 = add(x = var_8210_cast_fp16, y = var_8216_cast_fp16)[name = tensor("x_129_cast_fp16")]; + tensor var_8218_to_fp16 = const()[name = tensor("op_8218_to_fp16"), val = tensor(0x1p-1)]; + tensor var_8219_cast_fp16 = mul(x = x_129_cast_fp16, y = var_8218_to_fp16)[name = tensor("op_8219_cast_fp16")]; + tensor inputs_213_cast_fp16 = add(x = inputs_211_cast_fp16, y = var_8219_cast_fp16)[name = tensor("inputs_213_cast_fp16")]; + tensor out_213_axes_0 = const()[name = tensor("out_213_axes_0"), val = tensor([1])]; + tensor var_8229_to_fp16 = const()[name = tensor("op_8229_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_213_cast_fp16 = layer_norm(axes = out_213_axes_0, epsilon = var_8229_to_fp16, x = inputs_213_cast_fp16)[name = tensor("out_213_cast_fp16")]; + tensor obj_87_gamma_0_to_fp16 = const()[name = tensor("obj_87_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(495497920)))]; + tensor obj_87_beta_0_to_fp16 = const()[name = tensor("obj_87_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(495500032)))]; + tensor obj_87_epsilon_0_to_fp16 = const()[name = tensor("obj_87_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_87_cast_fp16 = batch_norm(beta = obj_87_beta_0_to_fp16, epsilon = obj_87_epsilon_0_to_fp16, gamma = obj_87_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_213_cast_fp16)[name = tensor("obj_87_cast_fp16")]; + tensor var_8254_pad_type_0 = const()[name = tensor("op_8254_pad_type_0"), val = tensor("valid")]; + tensor var_8254_strides_0 = const()[name = tensor("op_8254_strides_0"), val = tensor([1, 1])]; + tensor var_8254_pad_0 = const()[name = tensor("op_8254_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8254_dilations_0 = const()[name = tensor("op_8254_dilations_0"), val = tensor([1, 1])]; + tensor var_8254_groups_0 = const()[name = tensor("op_8254_groups_0"), val = tensor(1)]; + tensor layers_21_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(495502144))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496288640))), name = tensor("layers_21_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8254_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8254_dilations_0, groups = var_8254_groups_0, pad = var_8254_pad_0, pad_type = var_8254_pad_type_0, strides = var_8254_strides_0, weight = layers_21_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_87_cast_fp16)[name = tensor("op_8254_cast_fp16")]; + tensor var_8260_pad_type_0 = const()[name = tensor("op_8260_pad_type_0"), val = tensor("valid")]; + tensor var_8260_strides_0 = const()[name = tensor("op_8260_strides_0"), val = tensor([1, 1])]; + tensor var_8260_pad_0 = const()[name = tensor("op_8260_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8260_dilations_0 = const()[name = tensor("op_8260_dilations_0"), val = tensor([1, 1])]; + tensor var_8260_groups_0 = const()[name = tensor("op_8260_groups_0"), val = tensor(1)]; + tensor layers_21_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496321920))), name = tensor("layers_21_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496288832))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8260_cast_fp16 = conv(dilations = var_8260_dilations_0, groups = var_8260_groups_0, pad = var_8260_pad_0, pad_type = var_8260_pad_type_0, strides = var_8260_strides_0, weight = layers_21_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_87_cast_fp16)[name = tensor("op_8260_cast_fp16")]; + tensor query_85_cast_fp16 = add(x = var_8254_cast_fp16, y = var_8260_cast_fp16)[name = tensor("query_85_cast_fp16")]; + tensor var_8269_pad_type_0 = const()[name = tensor("op_8269_pad_type_0"), val = tensor("valid")]; + tensor var_8269_strides_0 = const()[name = tensor("op_8269_strides_0"), val = tensor([1, 1])]; + tensor var_8269_pad_0 = const()[name = tensor("op_8269_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8269_dilations_0 = const()[name = tensor("op_8269_dilations_0"), val = tensor([1, 1])]; + tensor var_8269_groups_0 = const()[name = tensor("op_8269_groups_0"), val = tensor(1)]; + tensor layers_21_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496453056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(497239552))), name = tensor("layers_21_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8269_cast_fp16 = conv(dilations = var_8269_dilations_0, groups = var_8269_groups_0, pad = var_8269_pad_0, pad_type = var_8269_pad_type_0, strides = var_8269_strides_0, weight = layers_21_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_87_cast_fp16)[name = tensor("op_8269_cast_fp16")]; + tensor var_8275_pad_type_0 = const()[name = tensor("op_8275_pad_type_0"), val = tensor("valid")]; + tensor var_8275_strides_0 = const()[name = tensor("op_8275_strides_0"), val = tensor([1, 1])]; + tensor var_8275_pad_0 = const()[name = tensor("op_8275_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8275_dilations_0 = const()[name = tensor("op_8275_dilations_0"), val = tensor([1, 1])]; + tensor var_8275_groups_0 = const()[name = tensor("op_8275_groups_0"), val = tensor(1)]; + tensor layers_21_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(497272448))), name = tensor("layers_21_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(497239744))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8275_cast_fp16 = conv(dilations = var_8275_dilations_0, groups = var_8275_groups_0, pad = var_8275_pad_0, pad_type = var_8275_pad_type_0, strides = var_8275_strides_0, weight = layers_21_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_87_cast_fp16)[name = tensor("op_8275_cast_fp16")]; + tensor key_43_cast_fp16 = add(x = var_8269_cast_fp16, y = var_8275_cast_fp16)[name = tensor("key_43_cast_fp16")]; + tensor var_8285_pad_type_0 = const()[name = tensor("op_8285_pad_type_0"), val = tensor("valid")]; + tensor var_8285_strides_0 = const()[name = tensor("op_8285_strides_0"), val = tensor([1, 1])]; + tensor var_8285_pad_0 = const()[name = tensor("op_8285_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8285_dilations_0 = const()[name = tensor("op_8285_dilations_0"), val = tensor([1, 1])]; + tensor var_8285_groups_0 = const()[name = tensor("op_8285_groups_0"), val = tensor(1)]; + tensor layers_21_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(497403584))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(498190080))), name = tensor("layers_21_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8285_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8285_dilations_0, groups = var_8285_groups_0, pad = var_8285_pad_0, pad_type = var_8285_pad_type_0, strides = var_8285_strides_0, weight = layers_21_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_87_cast_fp16)[name = tensor("op_8285_cast_fp16")]; + tensor var_8291_pad_type_0 = const()[name = tensor("op_8291_pad_type_0"), val = tensor("valid")]; + tensor var_8291_strides_0 = const()[name = tensor("op_8291_strides_0"), val = tensor([1, 1])]; + tensor var_8291_pad_0 = const()[name = tensor("op_8291_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8291_dilations_0 = const()[name = tensor("op_8291_dilations_0"), val = tensor([1, 1])]; + tensor var_8291_groups_0 = const()[name = tensor("op_8291_groups_0"), val = tensor(1)]; + tensor layers_21_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(498219584))), name = tensor("layers_21_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(498190272))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8291_cast_fp16 = conv(dilations = var_8291_dilations_0, groups = var_8291_groups_0, pad = var_8291_pad_0, pad_type = var_8291_pad_type_0, strides = var_8291_strides_0, weight = layers_21_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_87_cast_fp16)[name = tensor("op_8291_cast_fp16")]; + tensor value_43_cast_fp16 = add(x = var_8285_cast_fp16, y = var_8291_cast_fp16)[name = tensor("value_43_cast_fp16")]; + tensor var_8294_to_fp16 = const()[name = tensor("op_8294_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(498350720)))]; + tensor query_87_cast_fp16 = add(x = query_85_cast_fp16, y = var_8294_to_fp16)[name = tensor("query_87_cast_fp16")]; + tensor var_8297_to_fp16 = const()[name = tensor("op_8297_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(498352832)))]; + tensor q_with_bias_v_43_cast_fp16 = add(x = query_85_cast_fp16, y = var_8297_to_fp16)[name = tensor("q_with_bias_v_43_cast_fp16")]; + tensor var_8307_pad_type_0 = const()[name = tensor("op_8307_pad_type_0"), val = tensor("valid")]; + tensor var_8307_strides_0 = const()[name = tensor("op_8307_strides_0"), val = tensor([1, 1])]; + tensor var_8307_pad_0 = const()[name = tensor("op_8307_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8307_dilations_0 = const()[name = tensor("op_8307_dilations_0"), val = tensor([1, 1])]; + tensor var_8307_groups_0 = const()[name = tensor("op_8307_groups_0"), val = tensor(1)]; + tensor layers_21_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(498354944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499141440))), name = tensor("layers_21_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8307_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8307_dilations_0, groups = var_8307_groups_0, pad = var_8307_pad_0, pad_type = var_8307_pad_type_0, strides = var_8307_strides_0, weight = layers_21_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_8307_cast_fp16")]; + tensor var_8313_pad_type_0 = const()[name = tensor("op_8313_pad_type_0"), val = tensor("valid")]; + tensor var_8313_strides_0 = const()[name = tensor("op_8313_strides_0"), val = tensor([1, 1])]; + tensor var_8313_pad_0 = const()[name = tensor("op_8313_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8313_dilations_0 = const()[name = tensor("op_8313_dilations_0"), val = tensor([1, 1])]; + tensor var_8313_groups_0 = const()[name = tensor("op_8313_groups_0"), val = tensor(1)]; + tensor layers_21_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499223296))), name = tensor("layers_21_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499141632))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8313_cast_fp16 = conv(dilations = var_8313_dilations_0, groups = var_8313_groups_0, pad = var_8313_pad_0, pad_type = var_8313_pad_type_0, strides = var_8313_strides_0, weight = layers_21_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_8313_cast_fp16")]; + tensor p_43_cast_fp16 = add(x = var_8307_cast_fp16, y = var_8313_cast_fp16)[name = tensor("p_43_cast_fp16")]; + tensor var_8317 = const()[name = tensor("op_8317"), val = tensor([1, 8, 128, 188])]; + tensor var_8318_cast_fp16 = reshape(shape = var_8317, x = q_with_bias_v_43_cast_fp16)[name = tensor("op_8318_cast_fp16")]; + tensor var_8319 = const()[name = tensor("op_8319"), val = tensor([1, 8, 128, -1])]; + tensor var_8320_cast_fp16 = reshape(shape = var_8319, x = p_43_cast_fp16)[name = tensor("op_8320_cast_fp16")]; + tensor matrix_bd_169_transpose_x_0 = const()[name = tensor("matrix_bd_169_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_169_transpose_y_0 = const()[name = tensor("matrix_bd_169_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_169_cast_fp16 = matmul(transpose_x = matrix_bd_169_transpose_x_0, transpose_y = matrix_bd_169_transpose_y_0, x = var_8318_cast_fp16, y = var_8320_cast_fp16)[name = tensor("matrix_bd_169_cast_fp16")]; + tensor matrix_bd_171_pad_0 = const()[name = tensor("matrix_bd_171_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_171_mode_0 = const()[name = tensor("matrix_bd_171_mode_0"), val = tensor("constant")]; + tensor const_241_to_fp16 = const()[name = tensor("const_241_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_171_cast_fp16 = pad(constant_val = const_241_to_fp16, mode = matrix_bd_171_mode_0, pad = matrix_bd_171_pad_0, x = matrix_bd_169_cast_fp16)[name = tensor("matrix_bd_171_cast_fp16")]; + tensor var_8329 = const()[name = tensor("op_8329"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_173_cast_fp16 = reshape(shape = var_8329, x = matrix_bd_171_cast_fp16)[name = tensor("matrix_bd_173_cast_fp16")]; + tensor var_8333_begin_0 = const()[name = tensor("op_8333_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_8333_end_0 = const()[name = tensor("op_8333_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_8333_end_mask_0 = const()[name = tensor("op_8333_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_8333_cast_fp16 = slice_by_index(begin = var_8333_begin_0, end = var_8333_end_0, end_mask = var_8333_end_mask_0, x = matrix_bd_173_cast_fp16)[name = tensor("op_8333_cast_fp16")]; + tensor var_8334 = const()[name = tensor("op_8334"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_175_cast_fp16 = reshape(shape = var_8334, x = var_8333_cast_fp16)[name = tensor("matrix_bd_175_cast_fp16")]; + tensor var_8339_begin_0 = const()[name = tensor("op_8339_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8339_end_0 = const()[name = tensor("op_8339_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_8339_end_mask_0 = const()[name = tensor("op_8339_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8339_cast_fp16 = slice_by_index(begin = var_8339_begin_0, end = var_8339_end_0, end_mask = var_8339_end_mask_0, x = matrix_bd_175_cast_fp16)[name = tensor("op_8339_cast_fp16")]; + tensor var_8340_to_fp16 = const()[name = tensor("op_8340_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_43_cast_fp16 = mul(x = var_8339_cast_fp16, y = var_8340_to_fp16)[name = tensor("qk_mask_43_cast_fp16")]; + tensor var_8344 = const()[name = tensor("op_8344"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_43_cast_fp16 = reshape(shape = var_8344, x = query_87_cast_fp16)[name = tensor("mh_q_43_cast_fp16")]; + tensor var_8346_to_fp16 = const()[name = tensor("op_8346_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_8347_cast_fp16 = mul(x = mh_q_43_cast_fp16, y = var_8346_to_fp16)[name = tensor("op_8347_cast_fp16")]; + tensor var_8350 = const()[name = tensor("op_8350"), val = tensor([1, 8, 128, 188])]; + tensor var_8351_cast_fp16 = reshape(shape = var_8350, x = key_43_cast_fp16)[name = tensor("op_8351_cast_fp16")]; + tensor mh_w_85_transpose_x_0 = const()[name = tensor("mh_w_85_transpose_x_0"), val = tensor(true)]; + tensor mh_w_85_transpose_y_0 = const()[name = tensor("mh_w_85_transpose_y_0"), val = tensor(false)]; + tensor mh_w_85_cast_fp16 = matmul(transpose_x = mh_w_85_transpose_x_0, transpose_y = mh_w_85_transpose_y_0, x = var_8347_cast_fp16, y = var_8351_cast_fp16)[name = tensor("mh_w_85_cast_fp16")]; + tensor mh_w_87_cast_fp16 = add(x = mh_w_85_cast_fp16, y = qk_mask_43_cast_fp16)[name = tensor("mh_w_87_cast_fp16")]; + tensor var_8355_cast_fp16 = softmax(axis = var_8142, x = mh_w_87_cast_fp16)[name = tensor("op_8355_cast_fp16")]; + tensor var_8356 = const()[name = tensor("op_8356"), val = tensor([1, 8, 128, 188])]; + tensor var_8357_cast_fp16 = reshape(shape = var_8356, x = value_43_cast_fp16)[name = tensor("op_8357_cast_fp16")]; + tensor attn_43_transpose_x_0 = const()[name = tensor("attn_43_transpose_x_0"), val = tensor(false)]; + tensor attn_43_transpose_y_0 = const()[name = tensor("attn_43_transpose_y_0"), val = tensor(true)]; + tensor attn_43_cast_fp16 = matmul(transpose_x = attn_43_transpose_x_0, transpose_y = attn_43_transpose_y_0, x = var_8357_cast_fp16, y = var_8355_cast_fp16)[name = tensor("attn_43_cast_fp16")]; + tensor var_8360 = const()[name = tensor("op_8360"), val = tensor([1, 1024, 1, 188])]; + tensor input_569_cast_fp16 = reshape(shape = var_8360, x = attn_43_cast_fp16)[name = tensor("input_569_cast_fp16")]; + tensor var_8370_pad_type_0 = const()[name = tensor("op_8370_pad_type_0"), val = tensor("valid")]; + tensor var_8370_strides_0 = const()[name = tensor("op_8370_strides_0"), val = tensor([1, 1])]; + tensor var_8370_pad_0 = const()[name = tensor("op_8370_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8370_dilations_0 = const()[name = tensor("op_8370_dilations_0"), val = tensor([1, 1])]; + tensor var_8370_groups_0 = const()[name = tensor("op_8370_groups_0"), val = tensor(1)]; + tensor layers_21_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499354432))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(500140928))), name = tensor("layers_21_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8370_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8370_dilations_0, groups = var_8370_groups_0, pad = var_8370_pad_0, pad_type = var_8370_pad_type_0, strides = var_8370_strides_0, weight = layers_21_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_569_cast_fp16)[name = tensor("op_8370_cast_fp16")]; + tensor var_8376_pad_type_0 = const()[name = tensor("op_8376_pad_type_0"), val = tensor("valid")]; + tensor var_8376_strides_0 = const()[name = tensor("op_8376_strides_0"), val = tensor([1, 1])]; + tensor var_8376_pad_0 = const()[name = tensor("op_8376_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8376_dilations_0 = const()[name = tensor("op_8376_dilations_0"), val = tensor([1, 1])]; + tensor var_8376_groups_0 = const()[name = tensor("op_8376_groups_0"), val = tensor(1)]; + tensor layers_21_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(500170560))), name = tensor("layers_21_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(500141120))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8376_cast_fp16 = conv(dilations = var_8376_dilations_0, groups = var_8376_groups_0, pad = var_8376_pad_0, pad_type = var_8376_pad_type_0, strides = var_8376_strides_0, weight = layers_21_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_569_cast_fp16)[name = tensor("op_8376_cast_fp16")]; + tensor obj_89_cast_fp16 = add(x = var_8370_cast_fp16, y = var_8376_cast_fp16)[name = tensor("obj_89_cast_fp16")]; + tensor inputs_215_cast_fp16 = add(x = inputs_213_cast_fp16, y = obj_89_cast_fp16)[name = tensor("inputs_215_cast_fp16")]; + tensor out_215_axes_0 = const()[name = tensor("out_215_axes_0"), val = tensor([1])]; + tensor var_8387_to_fp16 = const()[name = tensor("op_8387_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_215_cast_fp16 = layer_norm(axes = out_215_axes_0, epsilon = var_8387_to_fp16, x = inputs_215_cast_fp16)[name = tensor("out_215_cast_fp16")]; + tensor input_571_gamma_0_to_fp16 = const()[name = tensor("input_571_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(500301696)))]; + tensor input_571_beta_0_to_fp16 = const()[name = tensor("input_571_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(500303808)))]; + tensor input_571_epsilon_0_to_fp16 = const()[name = tensor("input_571_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_571_cast_fp16 = batch_norm(beta = input_571_beta_0_to_fp16, epsilon = input_571_epsilon_0_to_fp16, gamma = input_571_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_215_cast_fp16)[name = tensor("input_571_cast_fp16")]; + tensor var_8408_pad_type_0 = const()[name = tensor("op_8408_pad_type_0"), val = tensor("valid")]; + tensor var_8408_strides_0 = const()[name = tensor("op_8408_strides_0"), val = tensor([1, 1])]; + tensor var_8408_pad_0 = const()[name = tensor("op_8408_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8408_dilations_0 = const()[name = tensor("op_8408_dilations_0"), val = tensor([1, 1])]; + tensor var_8408_groups_0 = const()[name = tensor("op_8408_groups_0"), val = tensor(1)]; + tensor layers_21_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(500305920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(501878848))), name = tensor("layers_21_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_8408_cast_fp16 = conv(dilations = var_8408_dilations_0, groups = var_8408_groups_0, pad = var_8408_pad_0, pad_type = var_8408_pad_type_0, strides = var_8408_strides_0, weight = layers_21_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_571_cast_fp16)[name = tensor("op_8408_cast_fp16")]; + tensor var_8414_pad_type_0 = const()[name = tensor("op_8414_pad_type_0"), val = tensor("valid")]; + tensor var_8414_strides_0 = const()[name = tensor("op_8414_strides_0"), val = tensor([1, 1])]; + tensor var_8414_pad_0 = const()[name = tensor("op_8414_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8414_dilations_0 = const()[name = tensor("op_8414_dilations_0"), val = tensor([1, 1])]; + tensor var_8414_groups_0 = const()[name = tensor("op_8414_groups_0"), val = tensor(1)]; + tensor layers_21_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(501945600))), name = tensor("layers_21_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(501879040))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_8414_cast_fp16 = conv(dilations = var_8414_dilations_0, groups = var_8414_groups_0, pad = var_8414_pad_0, pad_type = var_8414_pad_type_0, strides = var_8414_strides_0, weight = layers_21_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_571_cast_fp16)[name = tensor("op_8414_cast_fp16")]; + tensor input_573_cast_fp16 = add(x = var_8408_cast_fp16, y = var_8414_cast_fp16)[name = tensor("input_573_cast_fp16")]; + tensor input_575_split_num_splits_0 = const()[name = tensor("input_575_split_num_splits_0"), val = tensor(2)]; + tensor input_575_split_axis_0 = const()[name = tensor("input_575_split_axis_0"), val = tensor(1)]; + tensor input_575_split_cast_fp16_0, tensor input_575_split_cast_fp16_1 = split(axis = input_575_split_axis_0, num_splits = input_575_split_num_splits_0, x = input_573_cast_fp16)[name = tensor("input_575_split_cast_fp16")]; + tensor input_575_split_1_sigmoid_cast_fp16 = sigmoid(x = input_575_split_cast_fp16_1)[name = tensor("input_575_split_1_sigmoid_cast_fp16")]; + tensor input_575_cast_fp16 = mul(x = input_575_split_cast_fp16_0, y = input_575_split_1_sigmoid_cast_fp16)[name = tensor("input_575_cast_fp16")]; + tensor input_577_pad_type_0 = const()[name = tensor("input_577_pad_type_0"), val = tensor("custom")]; + tensor input_577_pad_0 = const()[name = tensor("input_577_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_577_groups_0 = const()[name = tensor("input_577_groups_0"), val = tensor(1024)]; + tensor input_577_strides_0 = const()[name = tensor("input_577_strides_0"), val = tensor([1, 1])]; + tensor input_577_dilations_0 = const()[name = tensor("input_577_dilations_0"), val = tensor([1, 1])]; + tensor const_310_to_fp16 = const()[name = tensor("const_310_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(502207808)))]; + tensor const_311_to_fp16 = const()[name = tensor("const_311_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(502226304)))]; + tensor input_579_cast_fp16 = conv(bias = const_311_to_fp16, dilations = input_577_dilations_0, groups = input_577_groups_0, pad = input_577_pad_0, pad_type = input_577_pad_type_0, strides = input_577_strides_0, weight = const_310_to_fp16, x = input_575_cast_fp16)[name = tensor("input_579_cast_fp16")]; + tensor input_581_cast_fp16 = silu(x = input_579_cast_fp16)[name = tensor("input_581_cast_fp16")]; + tensor var_8436_pad_type_0 = const()[name = tensor("op_8436_pad_type_0"), val = tensor("valid")]; + tensor var_8436_strides_0 = const()[name = tensor("op_8436_strides_0"), val = tensor([1, 1])]; + tensor var_8436_pad_0 = const()[name = tensor("op_8436_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8436_dilations_0 = const()[name = tensor("op_8436_dilations_0"), val = tensor([1, 1])]; + tensor var_8436_groups_0 = const()[name = tensor("op_8436_groups_0"), val = tensor(1)]; + tensor layers_21_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(502228416))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(503014912))), name = tensor("layers_21_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8436_cast_fp16 = conv(dilations = var_8436_dilations_0, groups = var_8436_groups_0, pad = var_8436_pad_0, pad_type = var_8436_pad_type_0, strides = var_8436_strides_0, weight = layers_21_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_581_cast_fp16)[name = tensor("op_8436_cast_fp16")]; + tensor var_8442_pad_type_0 = const()[name = tensor("op_8442_pad_type_0"), val = tensor("valid")]; + tensor var_8442_strides_0 = const()[name = tensor("op_8442_strides_0"), val = tensor([1, 1])]; + tensor var_8442_pad_0 = const()[name = tensor("op_8442_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8442_dilations_0 = const()[name = tensor("op_8442_dilations_0"), val = tensor([1, 1])]; + tensor var_8442_groups_0 = const()[name = tensor("op_8442_groups_0"), val = tensor(1)]; + tensor layers_21_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(503048128))), name = tensor("layers_21_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(503015104))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8442_cast_fp16 = conv(dilations = var_8442_dilations_0, groups = var_8442_groups_0, pad = var_8442_pad_0, pad_type = var_8442_pad_type_0, strides = var_8442_strides_0, weight = layers_21_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_581_cast_fp16)[name = tensor("op_8442_cast_fp16")]; + tensor x_131_cast_fp16 = add(x = var_8436_cast_fp16, y = var_8442_cast_fp16)[name = tensor("x_131_cast_fp16")]; + tensor inputs_217_cast_fp16 = add(x = inputs_215_cast_fp16, y = x_131_cast_fp16)[name = tensor("inputs_217_cast_fp16")]; + tensor out_217_axes_0 = const()[name = tensor("out_217_axes_0"), val = tensor([1])]; + tensor var_8453_to_fp16 = const()[name = tensor("op_8453_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_217_cast_fp16 = layer_norm(axes = out_217_axes_0, epsilon = var_8453_to_fp16, x = inputs_217_cast_fp16)[name = tensor("out_217_cast_fp16")]; + tensor input_583_gamma_0_to_fp16 = const()[name = tensor("input_583_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(503179264)))]; + tensor input_583_beta_0_to_fp16 = const()[name = tensor("input_583_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(503181376)))]; + tensor input_583_epsilon_0_to_fp16 = const()[name = tensor("input_583_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_583_cast_fp16 = batch_norm(beta = input_583_beta_0_to_fp16, epsilon = input_583_epsilon_0_to_fp16, gamma = input_583_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_217_cast_fp16)[name = tensor("input_583_cast_fp16")]; + tensor var_8473_pad_type_0 = const()[name = tensor("op_8473_pad_type_0"), val = tensor("valid")]; + tensor var_8473_strides_0 = const()[name = tensor("op_8473_strides_0"), val = tensor([1, 1])]; + tensor var_8473_pad_0 = const()[name = tensor("op_8473_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8473_dilations_0 = const()[name = tensor("op_8473_dilations_0"), val = tensor([1, 1])]; + tensor var_8473_groups_0 = const()[name = tensor("op_8473_groups_0"), val = tensor(1)]; + tensor layers_21_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(503183488))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(506329280))), name = tensor("layers_21_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_8473_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_8473_dilations_0, groups = var_8473_groups_0, pad = var_8473_pad_0, pad_type = var_8473_pad_type_0, strides = var_8473_strides_0, weight = layers_21_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_583_cast_fp16)[name = tensor("op_8473_cast_fp16")]; + tensor var_8479_pad_type_0 = const()[name = tensor("op_8479_pad_type_0"), val = tensor("valid")]; + tensor var_8479_strides_0 = const()[name = tensor("op_8479_strides_0"), val = tensor([1, 1])]; + tensor var_8479_pad_0 = const()[name = tensor("op_8479_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8479_dilations_0 = const()[name = tensor("op_8479_dilations_0"), val = tensor([1, 1])]; + tensor var_8479_groups_0 = const()[name = tensor("op_8479_groups_0"), val = tensor(1)]; + tensor layers_21_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(506460288))), name = tensor("layers_21_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(506329472))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_8479_cast_fp16 = conv(dilations = var_8479_dilations_0, groups = var_8479_groups_0, pad = var_8479_pad_0, pad_type = var_8479_pad_type_0, strides = var_8479_strides_0, weight = layers_21_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_583_cast_fp16)[name = tensor("op_8479_cast_fp16")]; + tensor input_585_cast_fp16 = add(x = var_8473_cast_fp16, y = var_8479_cast_fp16)[name = tensor("input_585_cast_fp16")]; + tensor input_587_cast_fp16 = silu(x = input_585_cast_fp16)[name = tensor("input_587_cast_fp16")]; + tensor var_8490_pad_type_0 = const()[name = tensor("op_8490_pad_type_0"), val = tensor("valid")]; + tensor var_8490_strides_0 = const()[name = tensor("op_8490_strides_0"), val = tensor([1, 1])]; + tensor var_8490_pad_0 = const()[name = tensor("op_8490_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8490_dilations_0 = const()[name = tensor("op_8490_dilations_0"), val = tensor([1, 1])]; + tensor var_8490_groups_0 = const()[name = tensor("op_8490_groups_0"), val = tensor(1)]; + tensor layers_21_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(506984640))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(510130432))), name = tensor("layers_21_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_8490_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8490_dilations_0, groups = var_8490_groups_0, pad = var_8490_pad_0, pad_type = var_8490_pad_type_0, strides = var_8490_strides_0, weight = layers_21_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_587_cast_fp16)[name = tensor("op_8490_cast_fp16")]; + tensor var_8496_pad_type_0 = const()[name = tensor("op_8496_pad_type_0"), val = tensor("valid")]; + tensor var_8496_strides_0 = const()[name = tensor("op_8496_strides_0"), val = tensor([1, 1])]; + tensor var_8496_pad_0 = const()[name = tensor("op_8496_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8496_dilations_0 = const()[name = tensor("op_8496_dilations_0"), val = tensor([1, 1])]; + tensor var_8496_groups_0 = const()[name = tensor("op_8496_groups_0"), val = tensor(1)]; + tensor layers_21_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(510288448))), name = tensor("layers_21_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(510130624))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_8496_cast_fp16 = conv(dilations = var_8496_dilations_0, groups = var_8496_groups_0, pad = var_8496_pad_0, pad_type = var_8496_pad_type_0, strides = var_8496_strides_0, weight = layers_21_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_587_cast_fp16)[name = tensor("op_8496_cast_fp16")]; + tensor x_133_cast_fp16 = add(x = var_8490_cast_fp16, y = var_8496_cast_fp16)[name = tensor("x_133_cast_fp16")]; + tensor var_8498_to_fp16 = const()[name = tensor("op_8498_to_fp16"), val = tensor(0x1p-1)]; + tensor var_8499_cast_fp16 = mul(x = x_133_cast_fp16, y = var_8498_to_fp16)[name = tensor("op_8499_cast_fp16")]; + tensor inputs_219_cast_fp16 = add(x = inputs_217_cast_fp16, y = var_8499_cast_fp16)[name = tensor("inputs_219_cast_fp16")]; + tensor out_219_axes_0 = const()[name = tensor("out_219_axes_0"), val = tensor([1])]; + tensor var_8509_to_fp16 = const()[name = tensor("op_8509_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_219_cast_fp16 = layer_norm(axes = out_219_axes_0, epsilon = var_8509_to_fp16, x = inputs_219_cast_fp16)[name = tensor("out_219_cast_fp16")]; + tensor inputs_221_gamma_0_to_fp16 = const()[name = tensor("inputs_221_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(510812800)))]; + tensor inputs_221_beta_0_to_fp16 = const()[name = tensor("inputs_221_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(510814912)))]; + tensor inputs_221_epsilon_0_to_fp16 = const()[name = tensor("inputs_221_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_221_cast_fp16 = batch_norm(beta = inputs_221_beta_0_to_fp16, epsilon = inputs_221_epsilon_0_to_fp16, gamma = inputs_221_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_219_cast_fp16)[name = tensor("inputs_221_cast_fp16")]; + tensor var_8523 = const()[name = tensor("op_8523"), val = tensor(3)]; + tensor out_221_axes_0 = const()[name = tensor("out_221_axes_0"), val = tensor([1])]; + tensor var_8554_to_fp16 = const()[name = tensor("op_8554_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_221_cast_fp16 = layer_norm(axes = out_221_axes_0, epsilon = var_8554_to_fp16, x = inputs_221_cast_fp16)[name = tensor("out_221_cast_fp16")]; + tensor input_589_gamma_0_to_fp16 = const()[name = tensor("input_589_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(510817024)))]; + tensor input_589_beta_0_to_fp16 = const()[name = tensor("input_589_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(510819136)))]; + tensor input_589_epsilon_0_to_fp16 = const()[name = tensor("input_589_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_589_cast_fp16 = batch_norm(beta = input_589_beta_0_to_fp16, epsilon = input_589_epsilon_0_to_fp16, gamma = input_589_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_221_cast_fp16)[name = tensor("input_589_cast_fp16")]; + tensor var_8574_pad_type_0 = const()[name = tensor("op_8574_pad_type_0"), val = tensor("valid")]; + tensor var_8574_strides_0 = const()[name = tensor("op_8574_strides_0"), val = tensor([1, 1])]; + tensor var_8574_pad_0 = const()[name = tensor("op_8574_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8574_dilations_0 = const()[name = tensor("op_8574_dilations_0"), val = tensor([1, 1])]; + tensor var_8574_groups_0 = const()[name = tensor("op_8574_groups_0"), val = tensor(1)]; + tensor layers_22_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(510821248))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(513967040))), name = tensor("layers_22_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_8574_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_8574_dilations_0, groups = var_8574_groups_0, pad = var_8574_pad_0, pad_type = var_8574_pad_type_0, strides = var_8574_strides_0, weight = layers_22_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_589_cast_fp16)[name = tensor("op_8574_cast_fp16")]; + tensor var_8580_pad_type_0 = const()[name = tensor("op_8580_pad_type_0"), val = tensor("valid")]; + tensor var_8580_strides_0 = const()[name = tensor("op_8580_strides_0"), val = tensor([1, 1])]; + tensor var_8580_pad_0 = const()[name = tensor("op_8580_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8580_dilations_0 = const()[name = tensor("op_8580_dilations_0"), val = tensor([1, 1])]; + tensor var_8580_groups_0 = const()[name = tensor("op_8580_groups_0"), val = tensor(1)]; + tensor layers_22_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(514095936))), name = tensor("layers_22_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(513967232))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_8580_cast_fp16 = conv(dilations = var_8580_dilations_0, groups = var_8580_groups_0, pad = var_8580_pad_0, pad_type = var_8580_pad_type_0, strides = var_8580_strides_0, weight = layers_22_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_589_cast_fp16)[name = tensor("op_8580_cast_fp16")]; + tensor input_591_cast_fp16 = add(x = var_8574_cast_fp16, y = var_8580_cast_fp16)[name = tensor("input_591_cast_fp16")]; + tensor input_593_cast_fp16 = silu(x = input_591_cast_fp16)[name = tensor("input_593_cast_fp16")]; + tensor var_8591_pad_type_0 = const()[name = tensor("op_8591_pad_type_0"), val = tensor("valid")]; + tensor var_8591_strides_0 = const()[name = tensor("op_8591_strides_0"), val = tensor([1, 1])]; + tensor var_8591_pad_0 = const()[name = tensor("op_8591_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8591_dilations_0 = const()[name = tensor("op_8591_dilations_0"), val = tensor([1, 1])]; + tensor var_8591_groups_0 = const()[name = tensor("op_8591_groups_0"), val = tensor(1)]; + tensor layers_22_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(514620288))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(517766080))), name = tensor("layers_22_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_8591_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8591_dilations_0, groups = var_8591_groups_0, pad = var_8591_pad_0, pad_type = var_8591_pad_type_0, strides = var_8591_strides_0, weight = layers_22_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_593_cast_fp16)[name = tensor("op_8591_cast_fp16")]; + tensor var_8597_pad_type_0 = const()[name = tensor("op_8597_pad_type_0"), val = tensor("valid")]; + tensor var_8597_strides_0 = const()[name = tensor("op_8597_strides_0"), val = tensor([1, 1])]; + tensor var_8597_pad_0 = const()[name = tensor("op_8597_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8597_dilations_0 = const()[name = tensor("op_8597_dilations_0"), val = tensor([1, 1])]; + tensor var_8597_groups_0 = const()[name = tensor("op_8597_groups_0"), val = tensor(1)]; + tensor layers_22_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(517931648))), name = tensor("layers_22_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(517766272))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_8597_cast_fp16 = conv(dilations = var_8597_dilations_0, groups = var_8597_groups_0, pad = var_8597_pad_0, pad_type = var_8597_pad_type_0, strides = var_8597_strides_0, weight = layers_22_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_593_cast_fp16)[name = tensor("op_8597_cast_fp16")]; + tensor x_135_cast_fp16 = add(x = var_8591_cast_fp16, y = var_8597_cast_fp16)[name = tensor("x_135_cast_fp16")]; + tensor var_8599_to_fp16 = const()[name = tensor("op_8599_to_fp16"), val = tensor(0x1p-1)]; + tensor var_8600_cast_fp16 = mul(x = x_135_cast_fp16, y = var_8599_to_fp16)[name = tensor("op_8600_cast_fp16")]; + tensor inputs_223_cast_fp16 = add(x = inputs_221_cast_fp16, y = var_8600_cast_fp16)[name = tensor("inputs_223_cast_fp16")]; + tensor out_223_axes_0 = const()[name = tensor("out_223_axes_0"), val = tensor([1])]; + tensor var_8610_to_fp16 = const()[name = tensor("op_8610_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_223_cast_fp16 = layer_norm(axes = out_223_axes_0, epsilon = var_8610_to_fp16, x = inputs_223_cast_fp16)[name = tensor("out_223_cast_fp16")]; + tensor obj_91_gamma_0_to_fp16 = const()[name = tensor("obj_91_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(518456000)))]; + tensor obj_91_beta_0_to_fp16 = const()[name = tensor("obj_91_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(518458112)))]; + tensor obj_91_epsilon_0_to_fp16 = const()[name = tensor("obj_91_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_91_cast_fp16 = batch_norm(beta = obj_91_beta_0_to_fp16, epsilon = obj_91_epsilon_0_to_fp16, gamma = obj_91_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_223_cast_fp16)[name = tensor("obj_91_cast_fp16")]; + tensor var_8635_pad_type_0 = const()[name = tensor("op_8635_pad_type_0"), val = tensor("valid")]; + tensor var_8635_strides_0 = const()[name = tensor("op_8635_strides_0"), val = tensor([1, 1])]; + tensor var_8635_pad_0 = const()[name = tensor("op_8635_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8635_dilations_0 = const()[name = tensor("op_8635_dilations_0"), val = tensor([1, 1])]; + tensor var_8635_groups_0 = const()[name = tensor("op_8635_groups_0"), val = tensor(1)]; + tensor layers_22_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(518460224))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(519246720))), name = tensor("layers_22_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8635_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8635_dilations_0, groups = var_8635_groups_0, pad = var_8635_pad_0, pad_type = var_8635_pad_type_0, strides = var_8635_strides_0, weight = layers_22_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_91_cast_fp16)[name = tensor("op_8635_cast_fp16")]; + tensor var_8641_pad_type_0 = const()[name = tensor("op_8641_pad_type_0"), val = tensor("valid")]; + tensor var_8641_strides_0 = const()[name = tensor("op_8641_strides_0"), val = tensor([1, 1])]; + tensor var_8641_pad_0 = const()[name = tensor("op_8641_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8641_dilations_0 = const()[name = tensor("op_8641_dilations_0"), val = tensor([1, 1])]; + tensor var_8641_groups_0 = const()[name = tensor("op_8641_groups_0"), val = tensor(1)]; + tensor layers_22_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(519277888))), name = tensor("layers_22_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(519246912))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8641_cast_fp16 = conv(dilations = var_8641_dilations_0, groups = var_8641_groups_0, pad = var_8641_pad_0, pad_type = var_8641_pad_type_0, strides = var_8641_strides_0, weight = layers_22_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_91_cast_fp16)[name = tensor("op_8641_cast_fp16")]; + tensor query_89_cast_fp16 = add(x = var_8635_cast_fp16, y = var_8641_cast_fp16)[name = tensor("query_89_cast_fp16")]; + tensor var_8650_pad_type_0 = const()[name = tensor("op_8650_pad_type_0"), val = tensor("valid")]; + tensor var_8650_strides_0 = const()[name = tensor("op_8650_strides_0"), val = tensor([1, 1])]; + tensor var_8650_pad_0 = const()[name = tensor("op_8650_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8650_dilations_0 = const()[name = tensor("op_8650_dilations_0"), val = tensor([1, 1])]; + tensor var_8650_groups_0 = const()[name = tensor("op_8650_groups_0"), val = tensor(1)]; + tensor layers_22_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(519409024))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(520195520))), name = tensor("layers_22_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8650_cast_fp16 = conv(dilations = var_8650_dilations_0, groups = var_8650_groups_0, pad = var_8650_pad_0, pad_type = var_8650_pad_type_0, strides = var_8650_strides_0, weight = layers_22_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_91_cast_fp16)[name = tensor("op_8650_cast_fp16")]; + tensor var_8656_pad_type_0 = const()[name = tensor("op_8656_pad_type_0"), val = tensor("valid")]; + tensor var_8656_strides_0 = const()[name = tensor("op_8656_strides_0"), val = tensor([1, 1])]; + tensor var_8656_pad_0 = const()[name = tensor("op_8656_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8656_dilations_0 = const()[name = tensor("op_8656_dilations_0"), val = tensor([1, 1])]; + tensor var_8656_groups_0 = const()[name = tensor("op_8656_groups_0"), val = tensor(1)]; + tensor layers_22_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(520227264))), name = tensor("layers_22_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(520195712))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8656_cast_fp16 = conv(dilations = var_8656_dilations_0, groups = var_8656_groups_0, pad = var_8656_pad_0, pad_type = var_8656_pad_type_0, strides = var_8656_strides_0, weight = layers_22_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_91_cast_fp16)[name = tensor("op_8656_cast_fp16")]; + tensor key_45_cast_fp16 = add(x = var_8650_cast_fp16, y = var_8656_cast_fp16)[name = tensor("key_45_cast_fp16")]; + tensor var_8666_pad_type_0 = const()[name = tensor("op_8666_pad_type_0"), val = tensor("valid")]; + tensor var_8666_strides_0 = const()[name = tensor("op_8666_strides_0"), val = tensor([1, 1])]; + tensor var_8666_pad_0 = const()[name = tensor("op_8666_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8666_dilations_0 = const()[name = tensor("op_8666_dilations_0"), val = tensor([1, 1])]; + tensor var_8666_groups_0 = const()[name = tensor("op_8666_groups_0"), val = tensor(1)]; + tensor layers_22_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(520358400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(521144896))), name = tensor("layers_22_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8666_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8666_dilations_0, groups = var_8666_groups_0, pad = var_8666_pad_0, pad_type = var_8666_pad_type_0, strides = var_8666_strides_0, weight = layers_22_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_91_cast_fp16)[name = tensor("op_8666_cast_fp16")]; + tensor var_8672_pad_type_0 = const()[name = tensor("op_8672_pad_type_0"), val = tensor("valid")]; + tensor var_8672_strides_0 = const()[name = tensor("op_8672_strides_0"), val = tensor([1, 1])]; + tensor var_8672_pad_0 = const()[name = tensor("op_8672_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8672_dilations_0 = const()[name = tensor("op_8672_dilations_0"), val = tensor([1, 1])]; + tensor var_8672_groups_0 = const()[name = tensor("op_8672_groups_0"), val = tensor(1)]; + tensor layers_22_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(521176128))), name = tensor("layers_22_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(521145088))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8672_cast_fp16 = conv(dilations = var_8672_dilations_0, groups = var_8672_groups_0, pad = var_8672_pad_0, pad_type = var_8672_pad_type_0, strides = var_8672_strides_0, weight = layers_22_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_91_cast_fp16)[name = tensor("op_8672_cast_fp16")]; + tensor value_45_cast_fp16 = add(x = var_8666_cast_fp16, y = var_8672_cast_fp16)[name = tensor("value_45_cast_fp16")]; + tensor var_8675_to_fp16 = const()[name = tensor("op_8675_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(521307264)))]; + tensor query_91_cast_fp16 = add(x = query_89_cast_fp16, y = var_8675_to_fp16)[name = tensor("query_91_cast_fp16")]; + tensor var_8678_to_fp16 = const()[name = tensor("op_8678_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(521309376)))]; + tensor q_with_bias_v_45_cast_fp16 = add(x = query_89_cast_fp16, y = var_8678_to_fp16)[name = tensor("q_with_bias_v_45_cast_fp16")]; + tensor var_8688_pad_type_0 = const()[name = tensor("op_8688_pad_type_0"), val = tensor("valid")]; + tensor var_8688_strides_0 = const()[name = tensor("op_8688_strides_0"), val = tensor([1, 1])]; + tensor var_8688_pad_0 = const()[name = tensor("op_8688_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8688_dilations_0 = const()[name = tensor("op_8688_dilations_0"), val = tensor([1, 1])]; + tensor var_8688_groups_0 = const()[name = tensor("op_8688_groups_0"), val = tensor(1)]; + tensor layers_22_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(521311488))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522097984))), name = tensor("layers_22_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8688_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8688_dilations_0, groups = var_8688_groups_0, pad = var_8688_pad_0, pad_type = var_8688_pad_type_0, strides = var_8688_strides_0, weight = layers_22_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_8688_cast_fp16")]; + tensor var_8694_pad_type_0 = const()[name = tensor("op_8694_pad_type_0"), val = tensor("valid")]; + tensor var_8694_strides_0 = const()[name = tensor("op_8694_strides_0"), val = tensor([1, 1])]; + tensor var_8694_pad_0 = const()[name = tensor("op_8694_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8694_dilations_0 = const()[name = tensor("op_8694_dilations_0"), val = tensor([1, 1])]; + tensor var_8694_groups_0 = const()[name = tensor("op_8694_groups_0"), val = tensor(1)]; + tensor layers_22_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522175424))), name = tensor("layers_22_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522098176))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8694_cast_fp16 = conv(dilations = var_8694_dilations_0, groups = var_8694_groups_0, pad = var_8694_pad_0, pad_type = var_8694_pad_type_0, strides = var_8694_strides_0, weight = layers_22_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_8694_cast_fp16")]; + tensor p_45_cast_fp16 = add(x = var_8688_cast_fp16, y = var_8694_cast_fp16)[name = tensor("p_45_cast_fp16")]; + tensor var_8698 = const()[name = tensor("op_8698"), val = tensor([1, 8, 128, 188])]; + tensor var_8699_cast_fp16 = reshape(shape = var_8698, x = q_with_bias_v_45_cast_fp16)[name = tensor("op_8699_cast_fp16")]; + tensor var_8700 = const()[name = tensor("op_8700"), val = tensor([1, 8, 128, -1])]; + tensor var_8701_cast_fp16 = reshape(shape = var_8700, x = p_45_cast_fp16)[name = tensor("op_8701_cast_fp16")]; + tensor matrix_bd_177_transpose_x_0 = const()[name = tensor("matrix_bd_177_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_177_transpose_y_0 = const()[name = tensor("matrix_bd_177_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_177_cast_fp16 = matmul(transpose_x = matrix_bd_177_transpose_x_0, transpose_y = matrix_bd_177_transpose_y_0, x = var_8699_cast_fp16, y = var_8701_cast_fp16)[name = tensor("matrix_bd_177_cast_fp16")]; + tensor matrix_bd_179_pad_0 = const()[name = tensor("matrix_bd_179_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_179_mode_0 = const()[name = tensor("matrix_bd_179_mode_0"), val = tensor("constant")]; + tensor const_252_to_fp16 = const()[name = tensor("const_252_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_179_cast_fp16 = pad(constant_val = const_252_to_fp16, mode = matrix_bd_179_mode_0, pad = matrix_bd_179_pad_0, x = matrix_bd_177_cast_fp16)[name = tensor("matrix_bd_179_cast_fp16")]; + tensor var_8710 = const()[name = tensor("op_8710"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_181_cast_fp16 = reshape(shape = var_8710, x = matrix_bd_179_cast_fp16)[name = tensor("matrix_bd_181_cast_fp16")]; + tensor var_8714_begin_0 = const()[name = tensor("op_8714_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_8714_end_0 = const()[name = tensor("op_8714_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_8714_end_mask_0 = const()[name = tensor("op_8714_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_8714_cast_fp16 = slice_by_index(begin = var_8714_begin_0, end = var_8714_end_0, end_mask = var_8714_end_mask_0, x = matrix_bd_181_cast_fp16)[name = tensor("op_8714_cast_fp16")]; + tensor var_8715 = const()[name = tensor("op_8715"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_183_cast_fp16 = reshape(shape = var_8715, x = var_8714_cast_fp16)[name = tensor("matrix_bd_183_cast_fp16")]; + tensor var_8720_begin_0 = const()[name = tensor("op_8720_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8720_end_0 = const()[name = tensor("op_8720_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_8720_end_mask_0 = const()[name = tensor("op_8720_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_8720_cast_fp16 = slice_by_index(begin = var_8720_begin_0, end = var_8720_end_0, end_mask = var_8720_end_mask_0, x = matrix_bd_183_cast_fp16)[name = tensor("op_8720_cast_fp16")]; + tensor var_8721_to_fp16 = const()[name = tensor("op_8721_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_45_cast_fp16 = mul(x = var_8720_cast_fp16, y = var_8721_to_fp16)[name = tensor("qk_mask_45_cast_fp16")]; + tensor var_8725 = const()[name = tensor("op_8725"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_45_cast_fp16 = reshape(shape = var_8725, x = query_91_cast_fp16)[name = tensor("mh_q_45_cast_fp16")]; + tensor var_8727_to_fp16 = const()[name = tensor("op_8727_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_8728_cast_fp16 = mul(x = mh_q_45_cast_fp16, y = var_8727_to_fp16)[name = tensor("op_8728_cast_fp16")]; + tensor var_8731 = const()[name = tensor("op_8731"), val = tensor([1, 8, 128, 188])]; + tensor var_8732_cast_fp16 = reshape(shape = var_8731, x = key_45_cast_fp16)[name = tensor("op_8732_cast_fp16")]; + tensor mh_w_89_transpose_x_0 = const()[name = tensor("mh_w_89_transpose_x_0"), val = tensor(true)]; + tensor mh_w_89_transpose_y_0 = const()[name = tensor("mh_w_89_transpose_y_0"), val = tensor(false)]; + tensor mh_w_89_cast_fp16 = matmul(transpose_x = mh_w_89_transpose_x_0, transpose_y = mh_w_89_transpose_y_0, x = var_8728_cast_fp16, y = var_8732_cast_fp16)[name = tensor("mh_w_89_cast_fp16")]; + tensor mh_w_91_cast_fp16 = add(x = mh_w_89_cast_fp16, y = qk_mask_45_cast_fp16)[name = tensor("mh_w_91_cast_fp16")]; + tensor var_8736_cast_fp16 = softmax(axis = var_8523, x = mh_w_91_cast_fp16)[name = tensor("op_8736_cast_fp16")]; + tensor var_8737 = const()[name = tensor("op_8737"), val = tensor([1, 8, 128, 188])]; + tensor var_8738_cast_fp16 = reshape(shape = var_8737, x = value_45_cast_fp16)[name = tensor("op_8738_cast_fp16")]; + tensor attn_45_transpose_x_0 = const()[name = tensor("attn_45_transpose_x_0"), val = tensor(false)]; + tensor attn_45_transpose_y_0 = const()[name = tensor("attn_45_transpose_y_0"), val = tensor(true)]; + tensor attn_45_cast_fp16 = matmul(transpose_x = attn_45_transpose_x_0, transpose_y = attn_45_transpose_y_0, x = var_8738_cast_fp16, y = var_8736_cast_fp16)[name = tensor("attn_45_cast_fp16")]; + tensor var_8741 = const()[name = tensor("op_8741"), val = tensor([1, 1024, 1, 188])]; + tensor input_595_cast_fp16 = reshape(shape = var_8741, x = attn_45_cast_fp16)[name = tensor("input_595_cast_fp16")]; + tensor var_8751_pad_type_0 = const()[name = tensor("op_8751_pad_type_0"), val = tensor("valid")]; + tensor var_8751_strides_0 = const()[name = tensor("op_8751_strides_0"), val = tensor([1, 1])]; + tensor var_8751_pad_0 = const()[name = tensor("op_8751_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8751_dilations_0 = const()[name = tensor("op_8751_dilations_0"), val = tensor([1, 1])]; + tensor var_8751_groups_0 = const()[name = tensor("op_8751_groups_0"), val = tensor(1)]; + tensor layers_22_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522306560))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(523093056))), name = tensor("layers_22_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8751_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8751_dilations_0, groups = var_8751_groups_0, pad = var_8751_pad_0, pad_type = var_8751_pad_type_0, strides = var_8751_strides_0, weight = layers_22_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_595_cast_fp16)[name = tensor("op_8751_cast_fp16")]; + tensor var_8757_pad_type_0 = const()[name = tensor("op_8757_pad_type_0"), val = tensor("valid")]; + tensor var_8757_strides_0 = const()[name = tensor("op_8757_strides_0"), val = tensor([1, 1])]; + tensor var_8757_pad_0 = const()[name = tensor("op_8757_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8757_dilations_0 = const()[name = tensor("op_8757_dilations_0"), val = tensor([1, 1])]; + tensor var_8757_groups_0 = const()[name = tensor("op_8757_groups_0"), val = tensor(1)]; + tensor layers_22_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(523123904))), name = tensor("layers_22_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(523093248))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8757_cast_fp16 = conv(dilations = var_8757_dilations_0, groups = var_8757_groups_0, pad = var_8757_pad_0, pad_type = var_8757_pad_type_0, strides = var_8757_strides_0, weight = layers_22_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_595_cast_fp16)[name = tensor("op_8757_cast_fp16")]; + tensor obj_93_cast_fp16 = add(x = var_8751_cast_fp16, y = var_8757_cast_fp16)[name = tensor("obj_93_cast_fp16")]; + tensor inputs_225_cast_fp16 = add(x = inputs_223_cast_fp16, y = obj_93_cast_fp16)[name = tensor("inputs_225_cast_fp16")]; + tensor out_225_axes_0 = const()[name = tensor("out_225_axes_0"), val = tensor([1])]; + tensor var_8768_to_fp16 = const()[name = tensor("op_8768_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_225_cast_fp16 = layer_norm(axes = out_225_axes_0, epsilon = var_8768_to_fp16, x = inputs_225_cast_fp16)[name = tensor("out_225_cast_fp16")]; + tensor input_597_gamma_0_to_fp16 = const()[name = tensor("input_597_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(523255040)))]; + tensor input_597_beta_0_to_fp16 = const()[name = tensor("input_597_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(523257152)))]; + tensor input_597_epsilon_0_to_fp16 = const()[name = tensor("input_597_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_597_cast_fp16 = batch_norm(beta = input_597_beta_0_to_fp16, epsilon = input_597_epsilon_0_to_fp16, gamma = input_597_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_225_cast_fp16)[name = tensor("input_597_cast_fp16")]; + tensor var_8789_pad_type_0 = const()[name = tensor("op_8789_pad_type_0"), val = tensor("valid")]; + tensor var_8789_strides_0 = const()[name = tensor("op_8789_strides_0"), val = tensor([1, 1])]; + tensor var_8789_pad_0 = const()[name = tensor("op_8789_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8789_dilations_0 = const()[name = tensor("op_8789_dilations_0"), val = tensor([1, 1])]; + tensor var_8789_groups_0 = const()[name = tensor("op_8789_groups_0"), val = tensor(1)]; + tensor layers_22_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(523259264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(524832192))), name = tensor("layers_22_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_8789_cast_fp16 = conv(dilations = var_8789_dilations_0, groups = var_8789_groups_0, pad = var_8789_pad_0, pad_type = var_8789_pad_type_0, strides = var_8789_strides_0, weight = layers_22_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_597_cast_fp16)[name = tensor("op_8789_cast_fp16")]; + tensor var_8795_pad_type_0 = const()[name = tensor("op_8795_pad_type_0"), val = tensor("valid")]; + tensor var_8795_strides_0 = const()[name = tensor("op_8795_strides_0"), val = tensor([1, 1])]; + tensor var_8795_pad_0 = const()[name = tensor("op_8795_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8795_dilations_0 = const()[name = tensor("op_8795_dilations_0"), val = tensor([1, 1])]; + tensor var_8795_groups_0 = const()[name = tensor("op_8795_groups_0"), val = tensor(1)]; + tensor layers_22_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(524905408))), name = tensor("layers_22_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(524832384))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_8795_cast_fp16 = conv(dilations = var_8795_dilations_0, groups = var_8795_groups_0, pad = var_8795_pad_0, pad_type = var_8795_pad_type_0, strides = var_8795_strides_0, weight = layers_22_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_597_cast_fp16)[name = tensor("op_8795_cast_fp16")]; + tensor input_599_cast_fp16 = add(x = var_8789_cast_fp16, y = var_8795_cast_fp16)[name = tensor("input_599_cast_fp16")]; + tensor input_601_split_num_splits_0 = const()[name = tensor("input_601_split_num_splits_0"), val = tensor(2)]; + tensor input_601_split_axis_0 = const()[name = tensor("input_601_split_axis_0"), val = tensor(1)]; + tensor input_601_split_cast_fp16_0, tensor input_601_split_cast_fp16_1 = split(axis = input_601_split_axis_0, num_splits = input_601_split_num_splits_0, x = input_599_cast_fp16)[name = tensor("input_601_split_cast_fp16")]; + tensor input_601_split_1_sigmoid_cast_fp16 = sigmoid(x = input_601_split_cast_fp16_1)[name = tensor("input_601_split_1_sigmoid_cast_fp16")]; + tensor input_601_cast_fp16 = mul(x = input_601_split_cast_fp16_0, y = input_601_split_1_sigmoid_cast_fp16)[name = tensor("input_601_cast_fp16")]; + tensor input_603_pad_type_0 = const()[name = tensor("input_603_pad_type_0"), val = tensor("custom")]; + tensor input_603_pad_0 = const()[name = tensor("input_603_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_603_groups_0 = const()[name = tensor("input_603_groups_0"), val = tensor(1024)]; + tensor input_603_strides_0 = const()[name = tensor("input_603_strides_0"), val = tensor([1, 1])]; + tensor input_603_dilations_0 = const()[name = tensor("input_603_dilations_0"), val = tensor([1, 1])]; + tensor const_312_to_fp16 = const()[name = tensor("const_312_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525167616)))]; + tensor const_313_to_fp16 = const()[name = tensor("const_313_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525186112)))]; + tensor input_605_cast_fp16 = conv(bias = const_313_to_fp16, dilations = input_603_dilations_0, groups = input_603_groups_0, pad = input_603_pad_0, pad_type = input_603_pad_type_0, strides = input_603_strides_0, weight = const_312_to_fp16, x = input_601_cast_fp16)[name = tensor("input_605_cast_fp16")]; + tensor input_607_cast_fp16 = silu(x = input_605_cast_fp16)[name = tensor("input_607_cast_fp16")]; + tensor var_8817_pad_type_0 = const()[name = tensor("op_8817_pad_type_0"), val = tensor("valid")]; + tensor var_8817_strides_0 = const()[name = tensor("op_8817_strides_0"), val = tensor([1, 1])]; + tensor var_8817_pad_0 = const()[name = tensor("op_8817_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8817_dilations_0 = const()[name = tensor("op_8817_dilations_0"), val = tensor([1, 1])]; + tensor var_8817_groups_0 = const()[name = tensor("op_8817_groups_0"), val = tensor(1)]; + tensor layers_22_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525188224))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525974720))), name = tensor("layers_22_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8817_cast_fp16 = conv(dilations = var_8817_dilations_0, groups = var_8817_groups_0, pad = var_8817_pad_0, pad_type = var_8817_pad_type_0, strides = var_8817_strides_0, weight = layers_22_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_607_cast_fp16)[name = tensor("op_8817_cast_fp16")]; + tensor var_8823_pad_type_0 = const()[name = tensor("op_8823_pad_type_0"), val = tensor("valid")]; + tensor var_8823_strides_0 = const()[name = tensor("op_8823_strides_0"), val = tensor([1, 1])]; + tensor var_8823_pad_0 = const()[name = tensor("op_8823_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8823_dilations_0 = const()[name = tensor("op_8823_dilations_0"), val = tensor([1, 1])]; + tensor var_8823_groups_0 = const()[name = tensor("op_8823_groups_0"), val = tensor(1)]; + tensor layers_22_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(526009024))), name = tensor("layers_22_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525974912))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_8823_cast_fp16 = conv(dilations = var_8823_dilations_0, groups = var_8823_groups_0, pad = var_8823_pad_0, pad_type = var_8823_pad_type_0, strides = var_8823_strides_0, weight = layers_22_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_607_cast_fp16)[name = tensor("op_8823_cast_fp16")]; + tensor x_137_cast_fp16 = add(x = var_8817_cast_fp16, y = var_8823_cast_fp16)[name = tensor("x_137_cast_fp16")]; + tensor inputs_227_cast_fp16 = add(x = inputs_225_cast_fp16, y = x_137_cast_fp16)[name = tensor("inputs_227_cast_fp16")]; + tensor out_227_axes_0 = const()[name = tensor("out_227_axes_0"), val = tensor([1])]; + tensor var_8834_to_fp16 = const()[name = tensor("op_8834_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_227_cast_fp16 = layer_norm(axes = out_227_axes_0, epsilon = var_8834_to_fp16, x = inputs_227_cast_fp16)[name = tensor("out_227_cast_fp16")]; + tensor input_609_gamma_0_to_fp16 = const()[name = tensor("input_609_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(526140160)))]; + tensor input_609_beta_0_to_fp16 = const()[name = tensor("input_609_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(526142272)))]; + tensor input_609_epsilon_0_to_fp16 = const()[name = tensor("input_609_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_609_cast_fp16 = batch_norm(beta = input_609_beta_0_to_fp16, epsilon = input_609_epsilon_0_to_fp16, gamma = input_609_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_227_cast_fp16)[name = tensor("input_609_cast_fp16")]; + tensor var_8854_pad_type_0 = const()[name = tensor("op_8854_pad_type_0"), val = tensor("valid")]; + tensor var_8854_strides_0 = const()[name = tensor("op_8854_strides_0"), val = tensor([1, 1])]; + tensor var_8854_pad_0 = const()[name = tensor("op_8854_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8854_dilations_0 = const()[name = tensor("op_8854_dilations_0"), val = tensor([1, 1])]; + tensor var_8854_groups_0 = const()[name = tensor("op_8854_groups_0"), val = tensor(1)]; + tensor layers_22_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(526144384))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529290176))), name = tensor("layers_22_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_8854_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_8854_dilations_0, groups = var_8854_groups_0, pad = var_8854_pad_0, pad_type = var_8854_pad_type_0, strides = var_8854_strides_0, weight = layers_22_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_609_cast_fp16)[name = tensor("op_8854_cast_fp16")]; + tensor var_8860_pad_type_0 = const()[name = tensor("op_8860_pad_type_0"), val = tensor("valid")]; + tensor var_8860_strides_0 = const()[name = tensor("op_8860_strides_0"), val = tensor([1, 1])]; + tensor var_8860_pad_0 = const()[name = tensor("op_8860_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8860_dilations_0 = const()[name = tensor("op_8860_dilations_0"), val = tensor([1, 1])]; + tensor var_8860_groups_0 = const()[name = tensor("op_8860_groups_0"), val = tensor(1)]; + tensor layers_22_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529428672))), name = tensor("layers_22_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529290368))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_8860_cast_fp16 = conv(dilations = var_8860_dilations_0, groups = var_8860_groups_0, pad = var_8860_pad_0, pad_type = var_8860_pad_type_0, strides = var_8860_strides_0, weight = layers_22_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_609_cast_fp16)[name = tensor("op_8860_cast_fp16")]; + tensor input_611_cast_fp16 = add(x = var_8854_cast_fp16, y = var_8860_cast_fp16)[name = tensor("input_611_cast_fp16")]; + tensor input_613_cast_fp16 = silu(x = input_611_cast_fp16)[name = tensor("input_613_cast_fp16")]; + tensor var_8871_pad_type_0 = const()[name = tensor("op_8871_pad_type_0"), val = tensor("valid")]; + tensor var_8871_strides_0 = const()[name = tensor("op_8871_strides_0"), val = tensor([1, 1])]; + tensor var_8871_pad_0 = const()[name = tensor("op_8871_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8871_dilations_0 = const()[name = tensor("op_8871_dilations_0"), val = tensor([1, 1])]; + tensor var_8871_groups_0 = const()[name = tensor("op_8871_groups_0"), val = tensor(1)]; + tensor layers_22_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529953024))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(533098816))), name = tensor("layers_22_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_8871_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8871_dilations_0, groups = var_8871_groups_0, pad = var_8871_pad_0, pad_type = var_8871_pad_type_0, strides = var_8871_strides_0, weight = layers_22_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_613_cast_fp16)[name = tensor("op_8871_cast_fp16")]; + tensor var_8877_pad_type_0 = const()[name = tensor("op_8877_pad_type_0"), val = tensor("valid")]; + tensor var_8877_strides_0 = const()[name = tensor("op_8877_strides_0"), val = tensor([1, 1])]; + tensor var_8877_pad_0 = const()[name = tensor("op_8877_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8877_dilations_0 = const()[name = tensor("op_8877_dilations_0"), val = tensor([1, 1])]; + tensor var_8877_groups_0 = const()[name = tensor("op_8877_groups_0"), val = tensor(1)]; + tensor layers_22_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(533319232))), name = tensor("layers_22_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(533099008))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_8877_cast_fp16 = conv(dilations = var_8877_dilations_0, groups = var_8877_groups_0, pad = var_8877_pad_0, pad_type = var_8877_pad_type_0, strides = var_8877_strides_0, weight = layers_22_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_613_cast_fp16)[name = tensor("op_8877_cast_fp16")]; + tensor x_139_cast_fp16 = add(x = var_8871_cast_fp16, y = var_8877_cast_fp16)[name = tensor("x_139_cast_fp16")]; + tensor var_8879_to_fp16 = const()[name = tensor("op_8879_to_fp16"), val = tensor(0x1p-1)]; + tensor var_8880_cast_fp16 = mul(x = x_139_cast_fp16, y = var_8879_to_fp16)[name = tensor("op_8880_cast_fp16")]; + tensor inputs_229_cast_fp16 = add(x = inputs_227_cast_fp16, y = var_8880_cast_fp16)[name = tensor("inputs_229_cast_fp16")]; + tensor out_229_axes_0 = const()[name = tensor("out_229_axes_0"), val = tensor([1])]; + tensor var_8890_to_fp16 = const()[name = tensor("op_8890_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_229_cast_fp16 = layer_norm(axes = out_229_axes_0, epsilon = var_8890_to_fp16, x = inputs_229_cast_fp16)[name = tensor("out_229_cast_fp16")]; + tensor inputs_231_gamma_0_to_fp16 = const()[name = tensor("inputs_231_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(533843584)))]; + tensor inputs_231_beta_0_to_fp16 = const()[name = tensor("inputs_231_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(533845696)))]; + tensor inputs_231_epsilon_0_to_fp16 = const()[name = tensor("inputs_231_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_231_cast_fp16 = batch_norm(beta = inputs_231_beta_0_to_fp16, epsilon = inputs_231_epsilon_0_to_fp16, gamma = inputs_231_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_229_cast_fp16)[name = tensor("inputs_231_cast_fp16")]; + tensor var_8904 = const()[name = tensor("op_8904"), val = tensor(3)]; + tensor out_231_axes_0 = const()[name = tensor("out_231_axes_0"), val = tensor([1])]; + tensor var_8935_to_fp16 = const()[name = tensor("op_8935_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_231_cast_fp16 = layer_norm(axes = out_231_axes_0, epsilon = var_8935_to_fp16, x = inputs_231_cast_fp16)[name = tensor("out_231_cast_fp16")]; + tensor input_615_gamma_0_to_fp16 = const()[name = tensor("input_615_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(533847808)))]; + tensor input_615_beta_0_to_fp16 = const()[name = tensor("input_615_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(533849920)))]; + tensor input_615_epsilon_0_to_fp16 = const()[name = tensor("input_615_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_615_cast_fp16 = batch_norm(beta = input_615_beta_0_to_fp16, epsilon = input_615_epsilon_0_to_fp16, gamma = input_615_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_231_cast_fp16)[name = tensor("input_615_cast_fp16")]; + tensor var_8955_pad_type_0 = const()[name = tensor("op_8955_pad_type_0"), val = tensor("valid")]; + tensor var_8955_strides_0 = const()[name = tensor("op_8955_strides_0"), val = tensor([1, 1])]; + tensor var_8955_pad_0 = const()[name = tensor("op_8955_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8955_dilations_0 = const()[name = tensor("op_8955_dilations_0"), val = tensor([1, 1])]; + tensor var_8955_groups_0 = const()[name = tensor("op_8955_groups_0"), val = tensor(1)]; + tensor layers_23_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(533852032))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(536997824))), name = tensor("layers_23_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_8955_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_8955_dilations_0, groups = var_8955_groups_0, pad = var_8955_pad_0, pad_type = var_8955_pad_type_0, strides = var_8955_strides_0, weight = layers_23_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_615_cast_fp16)[name = tensor("op_8955_cast_fp16")]; + tensor var_8961_pad_type_0 = const()[name = tensor("op_8961_pad_type_0"), val = tensor("valid")]; + tensor var_8961_strides_0 = const()[name = tensor("op_8961_strides_0"), val = tensor([1, 1])]; + tensor var_8961_pad_0 = const()[name = tensor("op_8961_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8961_dilations_0 = const()[name = tensor("op_8961_dilations_0"), val = tensor([1, 1])]; + tensor var_8961_groups_0 = const()[name = tensor("op_8961_groups_0"), val = tensor(1)]; + tensor layers_23_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(537124992))), name = tensor("layers_23_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(536998016))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_8961_cast_fp16 = conv(dilations = var_8961_dilations_0, groups = var_8961_groups_0, pad = var_8961_pad_0, pad_type = var_8961_pad_type_0, strides = var_8961_strides_0, weight = layers_23_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_615_cast_fp16)[name = tensor("op_8961_cast_fp16")]; + tensor input_617_cast_fp16 = add(x = var_8955_cast_fp16, y = var_8961_cast_fp16)[name = tensor("input_617_cast_fp16")]; + tensor input_619_cast_fp16 = silu(x = input_617_cast_fp16)[name = tensor("input_619_cast_fp16")]; + tensor var_8972_pad_type_0 = const()[name = tensor("op_8972_pad_type_0"), val = tensor("valid")]; + tensor var_8972_strides_0 = const()[name = tensor("op_8972_strides_0"), val = tensor([1, 1])]; + tensor var_8972_pad_0 = const()[name = tensor("op_8972_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8972_dilations_0 = const()[name = tensor("op_8972_dilations_0"), val = tensor([1, 1])]; + tensor var_8972_groups_0 = const()[name = tensor("op_8972_groups_0"), val = tensor(1)]; + tensor layers_23_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(537649344))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(540795136))), name = tensor("layers_23_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_8972_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_8972_dilations_0, groups = var_8972_groups_0, pad = var_8972_pad_0, pad_type = var_8972_pad_type_0, strides = var_8972_strides_0, weight = layers_23_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_619_cast_fp16)[name = tensor("op_8972_cast_fp16")]; + tensor var_8978_pad_type_0 = const()[name = tensor("op_8978_pad_type_0"), val = tensor("valid")]; + tensor var_8978_strides_0 = const()[name = tensor("op_8978_strides_0"), val = tensor([1, 1])]; + tensor var_8978_pad_0 = const()[name = tensor("op_8978_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_8978_dilations_0 = const()[name = tensor("op_8978_dilations_0"), val = tensor([1, 1])]; + tensor var_8978_groups_0 = const()[name = tensor("op_8978_groups_0"), val = tensor(1)]; + tensor layers_23_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(541007680))), name = tensor("layers_23_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(540795328))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_8978_cast_fp16 = conv(dilations = var_8978_dilations_0, groups = var_8978_groups_0, pad = var_8978_pad_0, pad_type = var_8978_pad_type_0, strides = var_8978_strides_0, weight = layers_23_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_619_cast_fp16)[name = tensor("op_8978_cast_fp16")]; + tensor x_141_cast_fp16 = add(x = var_8972_cast_fp16, y = var_8978_cast_fp16)[name = tensor("x_141_cast_fp16")]; + tensor var_8980_to_fp16 = const()[name = tensor("op_8980_to_fp16"), val = tensor(0x1p-1)]; + tensor var_8981_cast_fp16 = mul(x = x_141_cast_fp16, y = var_8980_to_fp16)[name = tensor("op_8981_cast_fp16")]; + tensor inputs_233_cast_fp16 = add(x = inputs_231_cast_fp16, y = var_8981_cast_fp16)[name = tensor("inputs_233_cast_fp16")]; + tensor out_233_axes_0 = const()[name = tensor("out_233_axes_0"), val = tensor([1])]; + tensor var_8991_to_fp16 = const()[name = tensor("op_8991_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_233_cast_fp16 = layer_norm(axes = out_233_axes_0, epsilon = var_8991_to_fp16, x = inputs_233_cast_fp16)[name = tensor("out_233_cast_fp16")]; + tensor obj_95_gamma_0_to_fp16 = const()[name = tensor("obj_95_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(541532032)))]; + tensor obj_95_beta_0_to_fp16 = const()[name = tensor("obj_95_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(541534144)))]; + tensor obj_95_epsilon_0_to_fp16 = const()[name = tensor("obj_95_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_95_cast_fp16 = batch_norm(beta = obj_95_beta_0_to_fp16, epsilon = obj_95_epsilon_0_to_fp16, gamma = obj_95_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_233_cast_fp16)[name = tensor("obj_95_cast_fp16")]; + tensor var_9016_pad_type_0 = const()[name = tensor("op_9016_pad_type_0"), val = tensor("valid")]; + tensor var_9016_strides_0 = const()[name = tensor("op_9016_strides_0"), val = tensor([1, 1])]; + tensor var_9016_pad_0 = const()[name = tensor("op_9016_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9016_dilations_0 = const()[name = tensor("op_9016_dilations_0"), val = tensor([1, 1])]; + tensor var_9016_groups_0 = const()[name = tensor("op_9016_groups_0"), val = tensor(1)]; + tensor layers_23_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(541536256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(542322752))), name = tensor("layers_23_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_9016_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_9016_dilations_0, groups = var_9016_groups_0, pad = var_9016_pad_0, pad_type = var_9016_pad_type_0, strides = var_9016_strides_0, weight = layers_23_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_95_cast_fp16)[name = tensor("op_9016_cast_fp16")]; + tensor var_9022_pad_type_0 = const()[name = tensor("op_9022_pad_type_0"), val = tensor("valid")]; + tensor var_9022_strides_0 = const()[name = tensor("op_9022_strides_0"), val = tensor([1, 1])]; + tensor var_9022_pad_0 = const()[name = tensor("op_9022_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9022_dilations_0 = const()[name = tensor("op_9022_dilations_0"), val = tensor([1, 1])]; + tensor var_9022_groups_0 = const()[name = tensor("op_9022_groups_0"), val = tensor(1)]; + tensor layers_23_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(542351808))), name = tensor("layers_23_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(542322944))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_9022_cast_fp16 = conv(dilations = var_9022_dilations_0, groups = var_9022_groups_0, pad = var_9022_pad_0, pad_type = var_9022_pad_type_0, strides = var_9022_strides_0, weight = layers_23_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_95_cast_fp16)[name = tensor("op_9022_cast_fp16")]; + tensor query_93_cast_fp16 = add(x = var_9016_cast_fp16, y = var_9022_cast_fp16)[name = tensor("query_93_cast_fp16")]; + tensor var_9031_pad_type_0 = const()[name = tensor("op_9031_pad_type_0"), val = tensor("valid")]; + tensor var_9031_strides_0 = const()[name = tensor("op_9031_strides_0"), val = tensor([1, 1])]; + tensor var_9031_pad_0 = const()[name = tensor("op_9031_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9031_dilations_0 = const()[name = tensor("op_9031_dilations_0"), val = tensor([1, 1])]; + tensor var_9031_groups_0 = const()[name = tensor("op_9031_groups_0"), val = tensor(1)]; + tensor layers_23_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(542482944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(543269440))), name = tensor("layers_23_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_9031_cast_fp16 = conv(dilations = var_9031_dilations_0, groups = var_9031_groups_0, pad = var_9031_pad_0, pad_type = var_9031_pad_type_0, strides = var_9031_strides_0, weight = layers_23_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_95_cast_fp16)[name = tensor("op_9031_cast_fp16")]; + tensor var_9037_pad_type_0 = const()[name = tensor("op_9037_pad_type_0"), val = tensor("valid")]; + tensor var_9037_strides_0 = const()[name = tensor("op_9037_strides_0"), val = tensor([1, 1])]; + tensor var_9037_pad_0 = const()[name = tensor("op_9037_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9037_dilations_0 = const()[name = tensor("op_9037_dilations_0"), val = tensor([1, 1])]; + tensor var_9037_groups_0 = const()[name = tensor("op_9037_groups_0"), val = tensor(1)]; + tensor layers_23_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(543299072))), name = tensor("layers_23_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(543269632))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_9037_cast_fp16 = conv(dilations = var_9037_dilations_0, groups = var_9037_groups_0, pad = var_9037_pad_0, pad_type = var_9037_pad_type_0, strides = var_9037_strides_0, weight = layers_23_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_95_cast_fp16)[name = tensor("op_9037_cast_fp16")]; + tensor key_cast_fp16 = add(x = var_9031_cast_fp16, y = var_9037_cast_fp16)[name = tensor("key_cast_fp16")]; + tensor var_9047_pad_type_0 = const()[name = tensor("op_9047_pad_type_0"), val = tensor("valid")]; + tensor var_9047_strides_0 = const()[name = tensor("op_9047_strides_0"), val = tensor([1, 1])]; + tensor var_9047_pad_0 = const()[name = tensor("op_9047_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9047_dilations_0 = const()[name = tensor("op_9047_dilations_0"), val = tensor([1, 1])]; + tensor var_9047_groups_0 = const()[name = tensor("op_9047_groups_0"), val = tensor(1)]; + tensor layers_23_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(543430208))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(544216704))), name = tensor("layers_23_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_9047_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_9047_dilations_0, groups = var_9047_groups_0, pad = var_9047_pad_0, pad_type = var_9047_pad_type_0, strides = var_9047_strides_0, weight = layers_23_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_95_cast_fp16)[name = tensor("op_9047_cast_fp16")]; + tensor var_9053_pad_type_0 = const()[name = tensor("op_9053_pad_type_0"), val = tensor("valid")]; + tensor var_9053_strides_0 = const()[name = tensor("op_9053_strides_0"), val = tensor([1, 1])]; + tensor var_9053_pad_0 = const()[name = tensor("op_9053_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9053_dilations_0 = const()[name = tensor("op_9053_dilations_0"), val = tensor([1, 1])]; + tensor var_9053_groups_0 = const()[name = tensor("op_9053_groups_0"), val = tensor(1)]; + tensor layers_23_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(544246720))), name = tensor("layers_23_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(544216896))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_9053_cast_fp16 = conv(dilations = var_9053_dilations_0, groups = var_9053_groups_0, pad = var_9053_pad_0, pad_type = var_9053_pad_type_0, strides = var_9053_strides_0, weight = layers_23_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_95_cast_fp16)[name = tensor("op_9053_cast_fp16")]; + tensor value_cast_fp16 = add(x = var_9047_cast_fp16, y = var_9053_cast_fp16)[name = tensor("value_cast_fp16")]; + tensor var_9056_to_fp16 = const()[name = tensor("op_9056_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(544377856)))]; + tensor query_cast_fp16 = add(x = query_93_cast_fp16, y = var_9056_to_fp16)[name = tensor("query_cast_fp16")]; + tensor var_9059_to_fp16 = const()[name = tensor("op_9059_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(544379968)))]; + tensor q_with_bias_v_cast_fp16 = add(x = query_93_cast_fp16, y = var_9059_to_fp16)[name = tensor("q_with_bias_v_cast_fp16")]; + tensor var_9069_pad_type_0 = const()[name = tensor("op_9069_pad_type_0"), val = tensor("valid")]; + tensor var_9069_strides_0 = const()[name = tensor("op_9069_strides_0"), val = tensor([1, 1])]; + tensor var_9069_pad_0 = const()[name = tensor("op_9069_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9069_dilations_0 = const()[name = tensor("op_9069_dilations_0"), val = tensor([1, 1])]; + tensor var_9069_groups_0 = const()[name = tensor("op_9069_groups_0"), val = tensor(1)]; + tensor layers_23_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(544382080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(545168576))), name = tensor("layers_23_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_9069_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_9069_dilations_0, groups = var_9069_groups_0, pad = var_9069_pad_0, pad_type = var_9069_pad_type_0, strides = var_9069_strides_0, weight = layers_23_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_9069_cast_fp16")]; + tensor var_9075_pad_type_0 = const()[name = tensor("op_9075_pad_type_0"), val = tensor("valid")]; + tensor var_9075_strides_0 = const()[name = tensor("op_9075_strides_0"), val = tensor([1, 1])]; + tensor var_9075_pad_0 = const()[name = tensor("op_9075_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9075_dilations_0 = const()[name = tensor("op_9075_dilations_0"), val = tensor([1, 1])]; + tensor var_9075_groups_0 = const()[name = tensor("op_9075_groups_0"), val = tensor(1)]; + tensor layers_23_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(545250176))), name = tensor("layers_23_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(545168768))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_9075_cast_fp16 = conv(dilations = var_9075_dilations_0, groups = var_9075_groups_0, pad = var_9075_pad_0, pad_type = var_9075_pad_type_0, strides = var_9075_strides_0, weight = layers_23_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_9075_cast_fp16")]; + tensor p_cast_fp16 = add(x = var_9069_cast_fp16, y = var_9075_cast_fp16)[name = tensor("p_cast_fp16")]; + tensor var_9079 = const()[name = tensor("op_9079"), val = tensor([1, 8, 128, 188])]; + tensor var_9080_cast_fp16 = reshape(shape = var_9079, x = q_with_bias_v_cast_fp16)[name = tensor("op_9080_cast_fp16")]; + tensor var_9081 = const()[name = tensor("op_9081"), val = tensor([1, 8, 128, -1])]; + tensor var_9082_cast_fp16 = reshape(shape = var_9081, x = p_cast_fp16)[name = tensor("op_9082_cast_fp16")]; + tensor matrix_bd_185_transpose_x_0 = const()[name = tensor("matrix_bd_185_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_185_transpose_y_0 = const()[name = tensor("matrix_bd_185_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_185_cast_fp16 = matmul(transpose_x = matrix_bd_185_transpose_x_0, transpose_y = matrix_bd_185_transpose_y_0, x = var_9080_cast_fp16, y = var_9082_cast_fp16)[name = tensor("matrix_bd_185_cast_fp16")]; + tensor matrix_bd_187_pad_0 = const()[name = tensor("matrix_bd_187_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_187_mode_0 = const()[name = tensor("matrix_bd_187_mode_0"), val = tensor("constant")]; + tensor const_263_to_fp16 = const()[name = tensor("const_263_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_187_cast_fp16 = pad(constant_val = const_263_to_fp16, mode = matrix_bd_187_mode_0, pad = matrix_bd_187_pad_0, x = matrix_bd_185_cast_fp16)[name = tensor("matrix_bd_187_cast_fp16")]; + tensor var_9091 = const()[name = tensor("op_9091"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_189_cast_fp16 = reshape(shape = var_9091, x = matrix_bd_187_cast_fp16)[name = tensor("matrix_bd_189_cast_fp16")]; + tensor var_9095_begin_0 = const()[name = tensor("op_9095_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_9095_end_0 = const()[name = tensor("op_9095_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_9095_end_mask_0 = const()[name = tensor("op_9095_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_9095_cast_fp16 = slice_by_index(begin = var_9095_begin_0, end = var_9095_end_0, end_mask = var_9095_end_mask_0, x = matrix_bd_189_cast_fp16)[name = tensor("op_9095_cast_fp16")]; + tensor var_9096 = const()[name = tensor("op_9096"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_cast_fp16 = reshape(shape = var_9096, x = var_9095_cast_fp16)[name = tensor("matrix_bd_cast_fp16")]; + tensor var_9101_begin_0 = const()[name = tensor("op_9101_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9101_end_0 = const()[name = tensor("op_9101_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_9101_end_mask_0 = const()[name = tensor("op_9101_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_9101_cast_fp16 = slice_by_index(begin = var_9101_begin_0, end = var_9101_end_0, end_mask = var_9101_end_mask_0, x = matrix_bd_cast_fp16)[name = tensor("op_9101_cast_fp16")]; + tensor var_9102_to_fp16 = const()[name = tensor("op_9102_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor qk_mask_cast_fp16 = mul(x = var_9101_cast_fp16, y = var_9102_to_fp16)[name = tensor("qk_mask_cast_fp16")]; + tensor var_9106 = const()[name = tensor("op_9106"), val = tensor([1, 8, 128, 188])]; + tensor mh_q_cast_fp16 = reshape(shape = var_9106, x = query_cast_fp16)[name = tensor("mh_q_cast_fp16")]; + tensor var_9108_to_fp16 = const()[name = tensor("op_9108_to_fp16"), val = tensor(0x1.6ap-4)]; + tensor var_9109_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_9108_to_fp16)[name = tensor("op_9109_cast_fp16")]; + tensor var_9112 = const()[name = tensor("op_9112"), val = tensor([1, 8, 128, 188])]; + tensor var_9113_cast_fp16 = reshape(shape = var_9112, x = key_cast_fp16)[name = tensor("op_9113_cast_fp16")]; + tensor mh_w_93_transpose_x_0 = const()[name = tensor("mh_w_93_transpose_x_0"), val = tensor(true)]; + tensor mh_w_93_transpose_y_0 = const()[name = tensor("mh_w_93_transpose_y_0"), val = tensor(false)]; + tensor mh_w_93_cast_fp16 = matmul(transpose_x = mh_w_93_transpose_x_0, transpose_y = mh_w_93_transpose_y_0, x = var_9109_cast_fp16, y = var_9113_cast_fp16)[name = tensor("mh_w_93_cast_fp16")]; + tensor mh_w_cast_fp16 = add(x = mh_w_93_cast_fp16, y = qk_mask_cast_fp16)[name = tensor("mh_w_cast_fp16")]; + tensor var_9117_cast_fp16 = softmax(axis = var_8904, x = mh_w_cast_fp16)[name = tensor("op_9117_cast_fp16")]; + tensor var_9118 = const()[name = tensor("op_9118"), val = tensor([1, 8, 128, 188])]; + tensor var_9119_cast_fp16 = reshape(shape = var_9118, x = value_cast_fp16)[name = tensor("op_9119_cast_fp16")]; + tensor attn_transpose_x_0 = const()[name = tensor("attn_transpose_x_0"), val = tensor(false)]; + tensor attn_transpose_y_0 = const()[name = tensor("attn_transpose_y_0"), val = tensor(true)]; + tensor attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_9119_cast_fp16, y = var_9117_cast_fp16)[name = tensor("attn_cast_fp16")]; + tensor var_9122 = const()[name = tensor("op_9122"), val = tensor([1, 1024, 1, 188])]; + tensor input_621_cast_fp16 = reshape(shape = var_9122, x = attn_cast_fp16)[name = tensor("input_621_cast_fp16")]; + tensor var_9132_pad_type_0 = const()[name = tensor("op_9132_pad_type_0"), val = tensor("valid")]; + tensor var_9132_strides_0 = const()[name = tensor("op_9132_strides_0"), val = tensor([1, 1])]; + tensor var_9132_pad_0 = const()[name = tensor("op_9132_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9132_dilations_0 = const()[name = tensor("op_9132_dilations_0"), val = tensor([1, 1])]; + tensor var_9132_groups_0 = const()[name = tensor("op_9132_groups_0"), val = tensor(1)]; + tensor layers_23_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(545381312))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(546167808))), name = tensor("layers_23_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_9132_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_9132_dilations_0, groups = var_9132_groups_0, pad = var_9132_pad_0, pad_type = var_9132_pad_type_0, strides = var_9132_strides_0, weight = layers_23_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_621_cast_fp16)[name = tensor("op_9132_cast_fp16")]; + tensor var_9138_pad_type_0 = const()[name = tensor("op_9138_pad_type_0"), val = tensor("valid")]; + tensor var_9138_strides_0 = const()[name = tensor("op_9138_strides_0"), val = tensor([1, 1])]; + tensor var_9138_pad_0 = const()[name = tensor("op_9138_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9138_dilations_0 = const()[name = tensor("op_9138_dilations_0"), val = tensor([1, 1])]; + tensor var_9138_groups_0 = const()[name = tensor("op_9138_groups_0"), val = tensor(1)]; + tensor layers_23_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(546201344))), name = tensor("layers_23_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(546168000))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_9138_cast_fp16 = conv(dilations = var_9138_dilations_0, groups = var_9138_groups_0, pad = var_9138_pad_0, pad_type = var_9138_pad_type_0, strides = var_9138_strides_0, weight = layers_23_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_621_cast_fp16)[name = tensor("op_9138_cast_fp16")]; + tensor obj_cast_fp16 = add(x = var_9132_cast_fp16, y = var_9138_cast_fp16)[name = tensor("obj_cast_fp16")]; + tensor inputs_235_cast_fp16 = add(x = inputs_233_cast_fp16, y = obj_cast_fp16)[name = tensor("inputs_235_cast_fp16")]; + tensor out_235_axes_0 = const()[name = tensor("out_235_axes_0"), val = tensor([1])]; + tensor var_9149_to_fp16 = const()[name = tensor("op_9149_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_235_cast_fp16 = layer_norm(axes = out_235_axes_0, epsilon = var_9149_to_fp16, x = inputs_235_cast_fp16)[name = tensor("out_235_cast_fp16")]; + tensor input_623_gamma_0_to_fp16 = const()[name = tensor("input_623_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(546332480)))]; + tensor input_623_beta_0_to_fp16 = const()[name = tensor("input_623_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(546334592)))]; + tensor input_623_epsilon_0_to_fp16 = const()[name = tensor("input_623_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_623_cast_fp16 = batch_norm(beta = input_623_beta_0_to_fp16, epsilon = input_623_epsilon_0_to_fp16, gamma = input_623_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_235_cast_fp16)[name = tensor("input_623_cast_fp16")]; + tensor var_9170_pad_type_0 = const()[name = tensor("op_9170_pad_type_0"), val = tensor("valid")]; + tensor var_9170_strides_0 = const()[name = tensor("op_9170_strides_0"), val = tensor([1, 1])]; + tensor var_9170_pad_0 = const()[name = tensor("op_9170_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9170_dilations_0 = const()[name = tensor("op_9170_dilations_0"), val = tensor([1, 1])]; + tensor var_9170_groups_0 = const()[name = tensor("op_9170_groups_0"), val = tensor(1)]; + tensor layers_23_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(546336704))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547909632))), name = tensor("layers_23_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 1024, 1, 1])]; + tensor var_9170_cast_fp16 = conv(dilations = var_9170_dilations_0, groups = var_9170_groups_0, pad = var_9170_pad_0, pad_type = var_9170_pad_type_0, strides = var_9170_strides_0, weight = layers_23_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_623_cast_fp16)[name = tensor("op_9170_cast_fp16")]; + tensor var_9176_pad_type_0 = const()[name = tensor("op_9176_pad_type_0"), val = tensor("valid")]; + tensor var_9176_strides_0 = const()[name = tensor("op_9176_strides_0"), val = tensor([1, 1])]; + tensor var_9176_pad_0 = const()[name = tensor("op_9176_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9176_dilations_0 = const()[name = tensor("op_9176_dilations_0"), val = tensor([1, 1])]; + tensor var_9176_groups_0 = const()[name = tensor("op_9176_groups_0"), val = tensor(1)]; + tensor layers_23_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547979456))), name = tensor("layers_23_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547909824))), shape = tensor([2048, 1024, 1, 1])]; + tensor var_9176_cast_fp16 = conv(dilations = var_9176_dilations_0, groups = var_9176_groups_0, pad = var_9176_pad_0, pad_type = var_9176_pad_type_0, strides = var_9176_strides_0, weight = layers_23_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_623_cast_fp16)[name = tensor("op_9176_cast_fp16")]; + tensor input_625_cast_fp16 = add(x = var_9170_cast_fp16, y = var_9176_cast_fp16)[name = tensor("input_625_cast_fp16")]; + tensor input_627_split_num_splits_0 = const()[name = tensor("input_627_split_num_splits_0"), val = tensor(2)]; + tensor input_627_split_axis_0 = const()[name = tensor("input_627_split_axis_0"), val = tensor(1)]; + tensor input_627_split_cast_fp16_0, tensor input_627_split_cast_fp16_1 = split(axis = input_627_split_axis_0, num_splits = input_627_split_num_splits_0, x = input_625_cast_fp16)[name = tensor("input_627_split_cast_fp16")]; + tensor input_627_split_1_sigmoid_cast_fp16 = sigmoid(x = input_627_split_cast_fp16_1)[name = tensor("input_627_split_1_sigmoid_cast_fp16")]; + tensor input_627_cast_fp16 = mul(x = input_627_split_cast_fp16_0, y = input_627_split_1_sigmoid_cast_fp16)[name = tensor("input_627_cast_fp16")]; + tensor input_629_pad_type_0 = const()[name = tensor("input_629_pad_type_0"), val = tensor("custom")]; + tensor input_629_pad_0 = const()[name = tensor("input_629_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_629_groups_0 = const()[name = tensor("input_629_groups_0"), val = tensor(1024)]; + tensor input_629_strides_0 = const()[name = tensor("input_629_strides_0"), val = tensor([1, 1])]; + tensor input_629_dilations_0 = const()[name = tensor("input_629_dilations_0"), val = tensor([1, 1])]; + tensor const_314_to_fp16 = const()[name = tensor("const_314_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(548241664)))]; + tensor const_315_to_fp16 = const()[name = tensor("const_315_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(548260160)))]; + tensor input_631_cast_fp16 = conv(bias = const_315_to_fp16, dilations = input_629_dilations_0, groups = input_629_groups_0, pad = input_629_pad_0, pad_type = input_629_pad_type_0, strides = input_629_strides_0, weight = const_314_to_fp16, x = input_627_cast_fp16)[name = tensor("input_631_cast_fp16")]; + tensor input_633_cast_fp16 = silu(x = input_631_cast_fp16)[name = tensor("input_633_cast_fp16")]; + tensor var_9198_pad_type_0 = const()[name = tensor("op_9198_pad_type_0"), val = tensor("valid")]; + tensor var_9198_strides_0 = const()[name = tensor("op_9198_strides_0"), val = tensor([1, 1])]; + tensor var_9198_pad_0 = const()[name = tensor("op_9198_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9198_dilations_0 = const()[name = tensor("op_9198_dilations_0"), val = tensor([1, 1])]; + tensor var_9198_groups_0 = const()[name = tensor("op_9198_groups_0"), val = tensor(1)]; + tensor layers_23_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(548262272))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(549048768))), name = tensor("layers_23_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 1024, 1, 1])]; + tensor var_9198_cast_fp16 = conv(dilations = var_9198_dilations_0, groups = var_9198_groups_0, pad = var_9198_pad_0, pad_type = var_9198_pad_type_0, strides = var_9198_strides_0, weight = layers_23_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_633_cast_fp16)[name = tensor("op_9198_cast_fp16")]; + tensor var_9204_pad_type_0 = const()[name = tensor("op_9204_pad_type_0"), val = tensor("valid")]; + tensor var_9204_strides_0 = const()[name = tensor("op_9204_strides_0"), val = tensor([1, 1])]; + tensor var_9204_pad_0 = const()[name = tensor("op_9204_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9204_dilations_0 = const()[name = tensor("op_9204_dilations_0"), val = tensor([1, 1])]; + tensor var_9204_groups_0 = const()[name = tensor("op_9204_groups_0"), val = tensor(1)]; + tensor layers_23_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(549085504))), name = tensor("layers_23_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(549048960))), shape = tensor([1024, 1024, 1, 1])]; + tensor var_9204_cast_fp16 = conv(dilations = var_9204_dilations_0, groups = var_9204_groups_0, pad = var_9204_pad_0, pad_type = var_9204_pad_type_0, strides = var_9204_strides_0, weight = layers_23_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_633_cast_fp16)[name = tensor("op_9204_cast_fp16")]; + tensor x_143_cast_fp16 = add(x = var_9198_cast_fp16, y = var_9204_cast_fp16)[name = tensor("x_143_cast_fp16")]; + tensor inputs_237_cast_fp16 = add(x = inputs_235_cast_fp16, y = x_143_cast_fp16)[name = tensor("inputs_237_cast_fp16")]; + tensor out_237_axes_0 = const()[name = tensor("out_237_axes_0"), val = tensor([1])]; + tensor var_9215_to_fp16 = const()[name = tensor("op_9215_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_237_cast_fp16 = layer_norm(axes = out_237_axes_0, epsilon = var_9215_to_fp16, x = inputs_237_cast_fp16)[name = tensor("out_237_cast_fp16")]; + tensor input_635_gamma_0_to_fp16 = const()[name = tensor("input_635_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(549216640)))]; + tensor input_635_beta_0_to_fp16 = const()[name = tensor("input_635_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(549218752)))]; + tensor input_635_epsilon_0_to_fp16 = const()[name = tensor("input_635_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_635_cast_fp16 = batch_norm(beta = input_635_beta_0_to_fp16, epsilon = input_635_epsilon_0_to_fp16, gamma = input_635_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_237_cast_fp16)[name = tensor("input_635_cast_fp16")]; + tensor var_9235_pad_type_0 = const()[name = tensor("op_9235_pad_type_0"), val = tensor("valid")]; + tensor var_9235_strides_0 = const()[name = tensor("op_9235_strides_0"), val = tensor([1, 1])]; + tensor var_9235_pad_0 = const()[name = tensor("op_9235_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9235_dilations_0 = const()[name = tensor("op_9235_dilations_0"), val = tensor([1, 1])]; + tensor var_9235_groups_0 = const()[name = tensor("op_9235_groups_0"), val = tensor(1)]; + tensor layers_23_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(549220864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552366656))), name = tensor("layers_23_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([4096, 1024, 1, 1])]; + tensor var_9235_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_9235_dilations_0, groups = var_9235_groups_0, pad = var_9235_pad_0, pad_type = var_9235_pad_type_0, strides = var_9235_strides_0, weight = layers_23_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_635_cast_fp16)[name = tensor("op_9235_cast_fp16")]; + tensor var_9241_pad_type_0 = const()[name = tensor("op_9241_pad_type_0"), val = tensor("valid")]; + tensor var_9241_strides_0 = const()[name = tensor("op_9241_strides_0"), val = tensor([1, 1])]; + tensor var_9241_pad_0 = const()[name = tensor("op_9241_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9241_dilations_0 = const()[name = tensor("op_9241_dilations_0"), val = tensor([1, 1])]; + tensor var_9241_groups_0 = const()[name = tensor("op_9241_groups_0"), val = tensor(1)]; + tensor layers_23_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552496512))), name = tensor("layers_23_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552366848))), shape = tensor([4096, 1024, 1, 1])]; + tensor var_9241_cast_fp16 = conv(dilations = var_9241_dilations_0, groups = var_9241_groups_0, pad = var_9241_pad_0, pad_type = var_9241_pad_type_0, strides = var_9241_strides_0, weight = layers_23_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_635_cast_fp16)[name = tensor("op_9241_cast_fp16")]; + tensor input_637_cast_fp16 = add(x = var_9235_cast_fp16, y = var_9241_cast_fp16)[name = tensor("input_637_cast_fp16")]; + tensor input_639_cast_fp16 = silu(x = input_637_cast_fp16)[name = tensor("input_639_cast_fp16")]; + tensor var_9252_pad_type_0 = const()[name = tensor("op_9252_pad_type_0"), val = tensor("valid")]; + tensor var_9252_strides_0 = const()[name = tensor("op_9252_strides_0"), val = tensor([1, 1])]; + tensor var_9252_pad_0 = const()[name = tensor("op_9252_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9252_dilations_0 = const()[name = tensor("op_9252_dilations_0"), val = tensor([1, 1])]; + tensor var_9252_groups_0 = const()[name = tensor("op_9252_groups_0"), val = tensor(1)]; + tensor layers_23_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(553020864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(556166656))), name = tensor("layers_23_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 4096, 1, 1])]; + tensor var_9252_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_9252_dilations_0, groups = var_9252_groups_0, pad = var_9252_pad_0, pad_type = var_9252_pad_type_0, strides = var_9252_strides_0, weight = layers_23_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_639_cast_fp16)[name = tensor("op_9252_cast_fp16")]; + tensor var_9258_pad_type_0 = const()[name = tensor("op_9258_pad_type_0"), val = tensor("valid")]; + tensor var_9258_strides_0 = const()[name = tensor("op_9258_strides_0"), val = tensor([1, 1])]; + tensor var_9258_pad_0 = const()[name = tensor("op_9258_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9258_dilations_0 = const()[name = tensor("op_9258_dilations_0"), val = tensor([1, 1])]; + tensor var_9258_groups_0 = const()[name = tensor("op_9258_groups_0"), val = tensor(1)]; + tensor layers_23_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(556344768))), name = tensor("layers_23_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(556166848))), shape = tensor([1024, 4096, 1, 1])]; + tensor var_9258_cast_fp16 = conv(dilations = var_9258_dilations_0, groups = var_9258_groups_0, pad = var_9258_pad_0, pad_type = var_9258_pad_type_0, strides = var_9258_strides_0, weight = layers_23_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_639_cast_fp16)[name = tensor("op_9258_cast_fp16")]; + tensor x_cast_fp16 = add(x = var_9252_cast_fp16, y = var_9258_cast_fp16)[name = tensor("x_cast_fp16")]; + tensor var_9260_to_fp16 = const()[name = tensor("op_9260_to_fp16"), val = tensor(0x1p-1)]; + tensor var_9261_cast_fp16 = mul(x = x_cast_fp16, y = var_9260_to_fp16)[name = tensor("op_9261_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_237_cast_fp16, y = var_9261_cast_fp16)[name = tensor("inputs_cast_fp16")]; + tensor out_239_axes_0 = const()[name = tensor("out_239_axes_0"), val = tensor([1])]; + tensor var_9271_to_fp16 = const()[name = tensor("op_9271_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_239_cast_fp16 = layer_norm(axes = out_239_axes_0, epsilon = var_9271_to_fp16, x = inputs_cast_fp16)[name = tensor("out_239_cast_fp16")]; + tensor input_gamma_0_to_fp16 = const()[name = tensor("input_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(556869120)))]; + tensor input_beta_0_to_fp16 = const()[name = tensor("input_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(556871232)))]; + tensor input_epsilon_0_to_fp16 = const()[name = tensor("input_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_cast_fp16 = batch_norm(beta = input_beta_0_to_fp16, epsilon = input_epsilon_0_to_fp16, gamma = input_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_239_cast_fp16)[name = tensor("input_cast_fp16")]; + tensor var_9291_pad_type_0 = const()[name = tensor("op_9291_pad_type_0"), val = tensor("valid")]; + tensor var_9291_strides_0 = const()[name = tensor("op_9291_strides_0"), val = tensor([1, 1])]; + tensor var_9291_pad_0 = const()[name = tensor("op_9291_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_9291_dilations_0 = const()[name = tensor("op_9291_dilations_0"), val = tensor([1, 1])]; + tensor var_9291_groups_0 = const()[name = tensor("op_9291_groups_0"), val = tensor(1)]; + tensor ctc_head_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(556873344))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(569457088))), name = tensor("ctc_head_weight_to_fp16_palettized"), shape = tensor([16385, 1024, 1, 1])]; + tensor ctc_head_bias_to_fp16 = const()[name = tensor("ctc_head_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(569457280)))]; + tensor ctc_head_raw_output = conv(bias = ctc_head_bias_to_fp16, dilations = var_9291_dilations_0, groups = var_9291_groups_0, pad = var_9291_pad_0, pad_type = var_9291_pad_type_0, strides = var_9291_strides_0, weight = ctc_head_weight_to_fp16_palettized, x = input_cast_fp16)[name = tensor("op_9291_cast_fp16")]; + } -> (ctc_head_raw_output); +} \ No newline at end of file