config.json

{
  "model_type": "pp_chart2table",
  "architectures": [
    "GotOcr2ForConditionalGeneration"
  ],
  "auto_map": {
    "AutoConfig": "configuration_pp_chart2table.PPChart2TableConfig",
    "AutoModel": "modeling_pp_chart2table.PPChart2TableForConditionalGeneration"
  },
  "freeze_vision_tower": false,
  "image_seq_length": 256,
  "use_im_start_end": true,
  "bos_token_id": 500,
  "eos_token_id": 500,
  "im_start_token": 151857,
  "im_end_token": 151858,
  "image_token_index": 151859,
  "vocab_size": 500,
  "initializer_range": 0.02,
  "max_window_layers": 21,
  "tie_word_embeddings": true,
  "dtype": "bfloat16",
  "use_cache": true,
  "use_sliding_window": false,
  "vision_config": {
    "architectures": [
      "PPChart2TableVisionModel"
    ],
    "auto_map": {
      "AutoConfig": "configuration_pp_chart2table.PPChart2TableConfig",
      "AutoModel": "modeling_pp_chart2table.PPChart2TableVisionModel"
    },
    "num_hidden_layers": 12,
    "hidden_size": 768,
    "output_channels": 256,
    "hidden_act": "gelu",
    "num_channels": 3,
    "image_size": 1024,
    "mlp_ratio": 4.0,
    "mlp_dim": 3072,
    "layer_norm_eps": 1e-05,
    "num_attention_heads": 12,
    "patch_size": 16,
    "qkv_bias": true,
    "use_rel_pos": true,
    "use_abs_pos": true,
    "global_attn_indexes": [
      2,
      5,
      8,
      11
    ],
    "window_size": 14,
    "attention_dropout": 0.0
  },
  "text_config": {
    "architectures": [
      "PPChart2TableTextModel"
    ],
    "auto_map": {
      "AutoConfig": "configuration_qwen2.Qwen2Config",
      "AutoModel": "modeling_qwen2.Qwen2Model"
    },
    "hidden_size": 1024,
    "hidden_act": "silu",
    "num_hidden_layers": 24,
    "num_attention_heads": 16,
    "num_key_value_heads": 16,
    "intermediate_size": 2816,
    "attention_dropout": 0.0,
    "sliding_window": 32768,
    "rms_norm_eps": 1e-06,
    "vocab_size": 151860,
    "max_position_embeddings": 32768,
    "rope_parameters": {
      "rope_theta": 1000000.0,
      "rope_type": "default"
    }
  }
}