config.json

{
  "architectures": [
    "Step3vForConditionalGeneration"
  ],
  "auto_map": {
    "AutoConfig": "stepfun-ai/step3--configuration_step3.Step3VLConfig",
    "AutoModelForCausalLM": "stepfun-ai/step3--modeling_step3.Step3vForConditionalGeneration"
  },
  "bos_token_id": 0,
  "eos_token_id": 128805,
  "hidden_size": 32,
  "im_end_token": "<im_end>",
  "im_patch_token": "<im_patch>",
  "im_start_token": "<im_start>",
  "image_token_id": 128001,
  "image_token_len": 169,
  "model_type": "step3_vl",
  "patch_token_len": 81,
  "projector_bias": false,
  "text_config": {
    "architectures": [
      "Step3TextForCausalLM"
    ],
    "head_dim": 256,
    "hidden_size": 32,
    "intermediate_size": 64,
    "max_position_embedding": 65536,
    "max_seq_len": 65536,
    "model_type": "step3_text",
    "moe_intermediate_size": 64,
    "moe_layers_enum": "1",
    "moe_num_experts": 8,
    "moe_top_k": 3,
    "norm_expert_weight": false,
    "num_attention_groups": 1,
    "num_attention_heads": 2,
    "num_hidden_layers": 2,
    "rms_norm_eps": 1e-05,
    "rope_scaling": null,
    "rope_theta": 500000,
    "share_expert_dim": 64,
    "share_q_dim": 512,
    "torch_dtype": "bfloat16",
    "vocab_size": 128815
  },
  "torch_dtype": "bfloat16",
  "transformers_version": "4.54.1",
  "understand_projector_stride": 2,
  "vision_config": {
    "hidden_act": "quick_gelu",
    "hidden_size": 64,
    "image_size": 728,
    "intermediate_size": 128,
    "layer_norm_eps": 1e-05,
    "model_type": "step3_vision_encoder",
    "num_attention_heads": 2,
    "num_channels": 3,
    "num_hidden_layers": 2,
    "output_hidden_size": 64,
    "patch_size": 14
  }
}