config.json

{
  "architectures": [ "GotOcr2VisionEncoder" ],
  "torch_dtype": "bfloat16", 
  "transformers_version": "4.49.0.dev0",
  "attention_dropout": 0.0,
  "global_attn_indexes": [
    2,
    5,
    8,
    11
  ],
  "hidden_act": "gelu",
  "hidden_size": 768,
  "image_size": 1024,
  "initializer_range": 1e-10,
  "layer_norm_eps": 1e-06,
  "mlp_dim": 3072,
  "num_attention_heads": 12,
  "num_channels": 3,
  "num_hidden_layers": 12,
  "output_channels": 256,
  "patch_size": 16,
  "qkv_bias": true,
  "transformers_version": "4.55.4",
  "use_abs_pos": true,
  "use_rel_pos": true,
  "window_size": 14
}