config.json

{
  "architectures": ["Qwen2_5_VLForConditionalGeneration"],
  "model_type": "qwen2_5_vl",
  "transformers_version": "4.49.0",
  "torch_dtype": "bfloat16",

  "processor_class": "OpenCUAProcessor",

  "hidden_act": "silu",
  "attention_dropout": 0.0,
  "initializer_range": 0.02,
  "rms_norm_eps": 1e-06,
  "tie_word_embeddings": false,
  "use_cache": true,

  "vocab_size": 152064,
  "max_position_embeddings": 128000,
  "sliding_window": 32768,
  "use_sliding_window": false,
  "max_window_layers": 64,

  "rope_scaling": { "type": "default" },
  "rope_theta": 1000000.0,

  "hidden_size": 5120,
  "intermediate_size": 27648,
  "num_hidden_layers": 64,
  "num_attention_heads": 40,
  "num_key_value_heads": 8,

  "bos_token_id": 151643,
  "eos_token_id": 151644,
  "pad_token_id": 152063,

  "vision_start_token_id": 151665,
  "vision_end_token_id": 151666,
  "vision_token_id": 151654,
  "image_token_id": 151667,
  "video_token_id": 151664,

  "vision_config": {
    "model_type": "qwen2_5_vl",
    "in_chans": 3,
    "hidden_size": 1280,
    "intermediate_size": 3456,
    "out_hidden_size": 5120,
    "spatial_patch_size": 14,
    "tokens_per_second": 2,
    "torch_dtype": "bfloat16"
  }
}