Upload canary-1b-v2

canary-1b-v2/AudioEncoder.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ea2bb79ff3927283c2cd29756ff5c9dc53036e6038ef53e22d93494e80f7df2f
+size 243

canary-1b-v2/AudioEncoder.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef5ed889fcb32e62f1c0d8c172fff4161bcb429d449a46834f3c7a2376d2d7be
+size 374

canary-1b-v2/AudioEncoder.mlmodelc/metadata.json

@@ -0,0 +1,87 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "storagePrecision" : "Mixed (Float16, Palettized (6 bits), Sparse)",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 16385 × 1 × 188)",
+        "shortDescription" : "",
+        "shape" : "[1, 16385, 1, 188]",
+        "name" : "ctc_head_raw_output",
+        "type" : "MultiArray"
+      }
+    ],
+    "modelParameters" : [
+
+    ],
+    "specificationVersion" : 8,
+    "mlProgramOperationTypeHistogram" : {
+      "Pad" : 24,
+      "Ios17.mul" : 121,
+      "Split" : 24,
+      "Ios17.transpose" : 1,
+      "Ios17.sub" : 1,
+      "Ios16.constexprLutToDense" : 266,
+      "Ios17.conv" : 560,
+      "Ios17.matmul" : 72,
+      "Ios16.sigmoid" : 24,
+      "Ios17.add" : 433,
+      "Ios17.sliceByIndex" : 48,
+      "Ios16.constexprSparseToDense" : 265,
+      "Ios16.relu" : 3,
+      "Ios17.batchNorm" : 120,
+      "Ios16.softmax" : 24,
+      "Ios17.reshape" : 193,
+      "Ios17.layerNorm" : 120,
+      "Ios16.silu" : 72
+    },
+    "computePrecision" : "Mixed (Float16, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "14.0",
+      "tvOS" : "17.0",
+      "visionOS" : "1.0",
+      "watchOS" : "10.0",
+      "iOS" : "17.0",
+      "macCatalyst" : "17.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.source_dialect" : "TorchScript",
+      "com.github.apple.coremltools.source" : "torch==2.5.0",
+      "com.github.apple.coremltools.version" : "8.3.0"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1 × 1501 × 128)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 1501, 128]",
+        "name" : "melspectrogram_features",
+        "type" : "MultiArray"
+      },
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1 × 1 × 1)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 1, 1]",
+        "name" : "input_1",
+        "type" : "MultiArray"
+      }
+    ],
+    "generatedClassName" : "AudioEncoder_mixedBitPalettized_6_bit_6_bit",
+    "method" : "predict"
+  }
+]

canary-1b-v2/AudioEncoder.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:804ea36c2fdf98a99af26e1314e20f7d5b3b1835409bbbde539d441e04d9b0d6
+size 569490114

canary-1b-v2/MelSpectrogram.mlmodelc/analytics/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:18a67d3ece5958dea8f020020a16d72c2d6763103e5a671a2641444167b1c38d
+size 243

canary-1b-v2/MelSpectrogram.mlmodelc/coremldata.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:319a6328b0dd0cade7f4249ed7bb68b50aa5ffde93e968fb796cf0f5ad735c9c
+size 331

canary-1b-v2/MelSpectrogram.mlmodelc/metadata.json

@@ -0,0 +1,77 @@
+[
+  {
+    "metadataOutputVersion" : "3.0",
+    "storagePrecision" : "Float32",
+    "outputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 1 × 1 × 1501 × 128)",
+        "shortDescription" : "",
+        "shape" : "[1, 1, 1501, 128]",
+        "name" : "melspectrogram_features",
+        "type" : "MultiArray"
+      }
+    ],
+    "modelParameters" : [
+
+    ],
+    "specificationVersion" : 8,
+    "mlProgramOperationTypeHistogram" : {
+      "Identity" : 1,
+      "Ios17.mul" : 2,
+      "Ios17.sqrt" : 1,
+      "Ios17.square" : 3,
+      "Ios17.transpose" : 1,
+      "Ios17.sub" : 2,
+      "Ios17.matmul" : 1,
+      "Ios17.conv" : 2,
+      "Ios17.log" : 1,
+      "Ios17.sliceByIndex" : 2,
+      "Ios17.add" : 3,
+      "Ios16.reduceMean" : 2,
+      "Ios17.realDiv" : 1,
+      "Ios17.expandDims" : 4,
+      "Ios17.squeeze" : 2,
+      "Ios17.reshape" : 2,
+      "Ios17.cast" : 2,
+      "Pad" : 2
+    },
+    "computePrecision" : "Mixed (Float16, Float32, Int32)",
+    "isUpdatable" : "0",
+    "stateSchema" : [
+
+    ],
+    "availability" : {
+      "macOS" : "14.0",
+      "tvOS" : "17.0",
+      "visionOS" : "1.0",
+      "watchOS" : "10.0",
+      "iOS" : "17.0",
+      "macCatalyst" : "17.0"
+    },
+    "modelType" : {
+      "name" : "MLModelType_mlProgram"
+    },
+    "userDefinedMetadata" : {
+      "com.github.apple.coremltools.source_dialect" : "TorchScript",
+      "com.github.apple.coremltools.source" : "torch==2.5.0",
+      "com.github.apple.coremltools.version" : "8.3.0"
+    },
+    "inputSchema" : [
+      {
+        "hasShapeFlexibility" : "0",
+        "isOptional" : "0",
+        "dataType" : "Float16",
+        "formattedType" : "MultiArray (Float16 240000)",
+        "shortDescription" : "",
+        "shape" : "[240000]",
+        "name" : "audio",
+        "type" : "MultiArray"
+      }
+    ],
+    "generatedClassName" : "MelSpectrogram",
+    "method" : "predict"
+  }
+]

canary-1b-v2/MelSpectrogram.mlmodelc/model.mil

@@ -0,0 +1,86 @@
+program(1.0)
+[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3404.16.1"}, {"coremlc-version", "3404.23.1"}, {"coremltools-component-torch", "2.5.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.3.0"}})]
+{
+    func main<ios17>(tensor<fp16, [240000]> audio) {
+            tensor<string, []> cast_0_dtype_0 = const()[name = tensor<string, []>("cast_0_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<fp32, [128, 257]> mel_filters = const()[name = tensor<string, []>("mel_filters"), val = tensor<fp32, [128, 257]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<int32, [1]> var_8_begin_0 = const()[name = tensor<string, []>("op_8_begin_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_8_end_0 = const()[name = tensor<string, []>("op_8_end_0"), val = tensor<int32, [1]>([240000])];
+            tensor<bool, [1]> var_8_end_mask_0 = const()[name = tensor<string, []>("op_8_end_mask_0"), val = tensor<bool, [1]>([true])];
+            tensor<fp32, [240000]> cast_0 = cast(dtype = cast_0_dtype_0, x = audio)[name = tensor<string, []>("cast_9")];
+            tensor<fp32, [239999]> var_8 = slice_by_index(begin = var_8_begin_0, end = var_8_end_0, end_mask = var_8_end_mask_0, x = cast_0)[name = tensor<string, []>("op_8")];
+            tensor<int32, [1]> var_13_begin_0 = const()[name = tensor<string, []>("op_13_begin_0"), val = tensor<int32, [1]>([0])];
+            tensor<int32, [1]> var_13_end_0 = const()[name = tensor<string, []>("op_13_end_0"), val = tensor<int32, [1]>([239999])];
+            tensor<bool, [1]> var_13_end_mask_0 = const()[name = tensor<string, []>("op_13_end_mask_0"), val = tensor<bool, [1]>([false])];
+            tensor<fp32, [239999]> var_13 = slice_by_index(begin = var_13_begin_0, end = var_13_end_0, end_mask = var_13_end_mask_0, x = cast_0)[name = tensor<string, []>("op_13")];
+            tensor<fp32, []> var_14 = const()[name = tensor<string, []>("op_14"), val = tensor<fp32, []>(0x1.f0a3d8p-1)];
+            tensor<fp32, [239999]> var_15 = mul(x = var_13, y = var_14)[name = tensor<string, []>("op_15")];
+            tensor<fp32, [239999]> input_1 = sub(x = var_8, y = var_15)[name = tensor<string, []>("input_1")];
+            tensor<fp32, []> const_0 = const()[name = tensor<string, []>("const_0"), val = tensor<fp32, []>(0x0p+0)];
+            tensor<int32, [2]> input_3_pad_0 = const()[name = tensor<string, []>("input_3_pad_0"), val = tensor<int32, [2]>([1, 0])];
+            tensor<string, []> input_3_mode_0 = const()[name = tensor<string, []>("input_3_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp32, [240000]> input_3 = pad(constant_val = const_0, mode = input_3_mode_0, pad = input_3_pad_0, x = input_1)[name = tensor<string, []>("input_3")];
+            tensor<int32, [3]> var_30 = const()[name = tensor<string, []>("op_30"), val = tensor<int32, [3]>([1, 1, 240000])];
+            tensor<fp32, [1, 1, 240000]> input_5 = reshape(shape = var_30, x = input_3)[name = tensor<string, []>("input_5")];
+            tensor<fp32, []> const_2 = const()[name = tensor<string, []>("const_2"), val = tensor<fp32, []>(0x0p+0)];
+            tensor<int32, [6]> input_7_pad_0 = const()[name = tensor<string, []>("input_7_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 256, 256])];
+            tensor<string, []> input_7_mode_0 = const()[name = tensor<string, []>("input_7_mode_0"), val = tensor<string, []>("reflect")];
+            tensor<fp32, [1, 1, 240512]> input_7 = pad(constant_val = const_2, mode = input_7_mode_0, pad = input_7_pad_0, x = input_5)[name = tensor<string, []>("input_7")];
+            tensor<int32, [1]> var_42 = const()[name = tensor<string, []>("op_42"), val = tensor<int32, [1]>([240512])];
+            tensor<fp32, [240512]> input = reshape(shape = var_42, x = input_7)[name = tensor<string, []>("input")];
+            tensor<int32, [1]> expand_dims_0_axes_0 = const()[name = tensor<string, []>("expand_dims_0_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp32, [1, 240512]> expand_dims_0 = expand_dims(axes = expand_dims_0_axes_0, x = input)[name = tensor<string, []>("expand_dims_0")];
+            tensor<fp32, [257, 1, 512]> expand_dims_1 = const()[name = tensor<string, []>("expand_dims_1"), val = tensor<fp32, [257, 1, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131712)))];
+            tensor<fp32, [257, 1, 512]> expand_dims_2 = const()[name = tensor<string, []>("expand_dims_2"), val = tensor<fp32, [257, 1, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(658112)))];
+            tensor<int32, [1]> expand_dims_3 = const()[name = tensor<string, []>("expand_dims_3"), val = tensor<int32, [1]>([160])];
+            tensor<int32, [1]> expand_dims_4_axes_0 = const()[name = tensor<string, []>("expand_dims_4_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp32, [1, 1, 240512]> expand_dims_4 = expand_dims(axes = expand_dims_4_axes_0, x = expand_dims_0)[name = tensor<string, []>("expand_dims_4")];
+            tensor<string, []> conv_0_pad_type_0 = const()[name = tensor<string, []>("conv_0_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [2]> conv_0_pad_0 = const()[name = tensor<string, []>("conv_0_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> conv_0_dilations_0 = const()[name = tensor<string, []>("conv_0_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> conv_0_groups_0 = const()[name = tensor<string, []>("conv_0_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 257, 1501]> conv_0 = conv(dilations = conv_0_dilations_0, groups = conv_0_groups_0, pad = conv_0_pad_0, pad_type = conv_0_pad_type_0, strides = expand_dims_3, weight = expand_dims_1, x = expand_dims_4)[name = tensor<string, []>("conv_0")];
+            tensor<string, []> conv_1_pad_type_0 = const()[name = tensor<string, []>("conv_1_pad_type_0"), val = tensor<string, []>("valid")];
+            tensor<int32, [2]> conv_1_pad_0 = const()[name = tensor<string, []>("conv_1_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<int32, [1]> conv_1_dilations_0 = const()[name = tensor<string, []>("conv_1_dilations_0"), val = tensor<int32, [1]>([1])];
+            tensor<int32, []> conv_1_groups_0 = const()[name = tensor<string, []>("conv_1_groups_0"), val = tensor<int32, []>(1)];
+            tensor<fp32, [1, 257, 1501]> conv_1 = conv(dilations = conv_1_dilations_0, groups = conv_1_groups_0, pad = conv_1_pad_0, pad_type = conv_1_pad_type_0, strides = expand_dims_3, weight = expand_dims_2, x = expand_dims_4)[name = tensor<string, []>("conv_1")];
+            tensor<int32, [1]> squeeze_0_axes_0 = const()[name = tensor<string, []>("squeeze_0_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp32, [257, 1501]> squeeze_0 = squeeze(axes = squeeze_0_axes_0, x = conv_0)[name = tensor<string, []>("squeeze_0")];
+            tensor<int32, [1]> squeeze_1_axes_0 = const()[name = tensor<string, []>("squeeze_1_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp32, [257, 1501]> squeeze_1 = squeeze(axes = squeeze_1_axes_0, x = conv_1)[name = tensor<string, []>("squeeze_1")];
+            tensor<fp32, [257, 1501]> square_1 = square(x = squeeze_0)[name = tensor<string, []>("square_1")];
+            tensor<fp32, [257, 1501]> square_2 = square(x = squeeze_1)[name = tensor<string, []>("square_2")];
+            tensor<fp32, [257, 1501]> add_1 = add(x = square_1, y = square_2)[name = tensor<string, []>("add_1")];
+            tensor<fp32, [257, 1501]> magnitudes = identity(x = add_1)[name = tensor<string, []>("magnitudes")];
+            tensor<bool, []> mel_spec_1_transpose_x_0 = const()[name = tensor<string, []>("mel_spec_1_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> mel_spec_1_transpose_y_0 = const()[name = tensor<string, []>("mel_spec_1_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp32, [128, 1501]> mel_spec_1 = matmul(transpose_x = mel_spec_1_transpose_x_0, transpose_y = mel_spec_1_transpose_y_0, x = mel_filters, y = magnitudes)[name = tensor<string, []>("mel_spec_1")];
+            tensor<fp32, []> var_56 = const()[name = tensor<string, []>("op_56"), val = tensor<fp32, []>(0x1p-24)];
+            tensor<fp32, [128, 1501]> mel_spec_3 = add(x = mel_spec_1, y = var_56)[name = tensor<string, []>("mel_spec_3")];
+            tensor<fp32, []> mel_spec_5_epsilon_0 = const()[name = tensor<string, []>("mel_spec_5_epsilon_0"), val = tensor<fp32, []>(0x1p-149)];
+            tensor<fp32, [128, 1501]> mel_spec_5 = log(epsilon = mel_spec_5_epsilon_0, x = mel_spec_3)[name = tensor<string, []>("mel_spec_5")];
+            tensor<int32, [1]> per_feature_mean_axes_0 = const()[name = tensor<string, []>("per_feature_mean_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<bool, []> per_feature_mean_keep_dims_0 = const()[name = tensor<string, []>("per_feature_mean_keep_dims_0"), val = tensor<bool, []>(true)];
+            tensor<fp32, [128, 1]> per_feature_mean = reduce_mean(axes = per_feature_mean_axes_0, keep_dims = per_feature_mean_keep_dims_0, x = mel_spec_5)[name = tensor<string, []>("per_feature_mean")];
+            tensor<fp32, [128, 1501]> sub_0 = sub(x = mel_spec_5, y = per_feature_mean)[name = tensor<string, []>("sub_0")];
+            tensor<fp32, [128, 1501]> square_0 = square(x = sub_0)[name = tensor<string, []>("square_0")];
+            tensor<int32, [1]> reduce_mean_1_axes_0 = const()[name = tensor<string, []>("reduce_mean_1_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<bool, []> reduce_mean_1_keep_dims_0 = const()[name = tensor<string, []>("reduce_mean_1_keep_dims_0"), val = tensor<bool, []>(true)];
+            tensor<fp32, [128, 1]> reduce_mean_1 = reduce_mean(axes = reduce_mean_1_axes_0, keep_dims = reduce_mean_1_keep_dims_0, x = square_0)[name = tensor<string, []>("reduce_mean_1")];
+            tensor<fp32, []> real_div_0 = const()[name = tensor<string, []>("real_div_0"), val = tensor<fp32, []>(0x1.002bbp+0)];
+            tensor<fp32, [128, 1]> mul_0 = mul(x = reduce_mean_1, y = real_div_0)[name = tensor<string, []>("mul_0")];
+            tensor<fp32, [128, 1]> sqrt_0 = sqrt(x = mul_0)[name = tensor<string, []>("sqrt_0")];
+            tensor<fp32, []> var_70 = const()[name = tensor<string, []>("op_70"), val = tensor<fp32, []>(0x1.4f8b58p-17)];
+            tensor<fp32, [128, 1]> per_feature_std = add(x = sqrt_0, y = var_70)[name = tensor<string, []>("per_feature_std")];
+            tensor<fp32, [128, 1501]> mel_spec = real_div(x = sub_0, y = per_feature_std)[name = tensor<string, []>("mel_spec")];
+            tensor<int32, [2]> var_75_perm_0 = const()[name = tensor<string, []>("op_75_perm_0"), val = tensor<int32, [2]>([1, 0])];
+            tensor<int32, [1]> var_77_axes_0 = const()[name = tensor<string, []>("op_77_axes_0"), val = tensor<int32, [1]>([0])];
+            tensor<fp32, [1501, 128]> var_75 = transpose(perm = var_75_perm_0, x = mel_spec)[name = tensor<string, []>("transpose_0")];
+            tensor<fp32, [1, 1501, 128]> var_77 = expand_dims(axes = var_77_axes_0, x = var_75)[name = tensor<string, []>("op_77")];
+            tensor<int32, [1]> var_79_axes_0 = const()[name = tensor<string, []>("op_79_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp32, [1, 1, 1501, 128]> var_79 = expand_dims(axes = var_79_axes_0, x = var_77)[name = tensor<string, []>("op_79")];
+            tensor<string, []> cast_7_dtype_0 = const()[name = tensor<string, []>("cast_7_dtype_0"), val = tensor<string, []>("fp16")];
+            tensor<fp16, [1, 1, 1501, 128]> melspectrogram_features = cast(dtype = cast_7_dtype_0, x = var_79)[name = tensor<string, []>("cast_8")];
+        } -> (melspectrogram_features);
+}

canary-1b-v2/MelSpectrogram.mlmodelc/weights/weight.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1570ad9beb3077cf61314606d8e4c30fc1d7b971e3bb73ee9444467faa9dd671
+size 1184512

canary-1b-v2/config.json

@@ -0,0 +1,37 @@
+{
+  "architectures": [
+    "ParakeetForCTC"
+  ],
+  "ctc_loss_reduction": "mean",
+  "ctc_zero_infinity": true,
+  "dtype": "float32",
+  "encoder_config": {
+    "activation_dropout": 0.1,
+    "attention_bias": true,
+    "attention_dropout": 0.1,
+    "conv_kernel_size": 9,
+    "dropout": 0.1,
+    "dropout_positions": 0.0,
+    "hidden_act": "silu",
+    "hidden_size": 1024,
+    "initializer_range": 0.02,
+    "intermediate_size": 4096,
+    "layerdrop": 0.1,
+    "max_position_embeddings": 5000,
+    "model_type": "parakeet_encoder",
+    "num_attention_heads": 8,
+    "num_hidden_layers": 24,
+    "num_key_value_heads": 8,
+    "num_mel_bins": 128,
+    "scale_input": false,
+    "subsampling_conv_channels": 256,
+    "subsampling_conv_kernel_size": 3,
+    "subsampling_conv_stride": 2,
+    "subsampling_factor": 8
+  },
+  "initializer_range": 0.02,
+  "model_type": "parakeet_ctc",
+  "pad_token_id": 1024,
+  "transformers_version": "4.57.1",
+  "vocab_size": 16385
+}

canary-1b-v2/preprocessor_config.json

@@ -0,0 +1,13 @@
+{
+  "feature_extractor_type": "ParakeetFeatureExtractor",
+  "feature_size": 128,
+  "hop_length": 160,
+  "n_fft": 512,
+  "padding_side": "right",
+  "padding_value": 0.0,
+  "preemphasis": 0.97,
+  "processor_class": "ParakeetProcessor",
+  "return_attention_mask": true,
+  "sampling_rate": 16000,
+  "win_length": 400
+}