config.json

{
  "architectures": [
    "F2PDecoderModel"
  ],
  "attention_probs_dropout_prob": 0.0,
  "auto_map": {
    "AutoConfig": "configuration_f2p_decoder.F2PDecoderConfig",
    "AutoModel": "modeling_f2p_decoder.F2PDecoderModel"
  },
  "decoder_hidden_size": 1152,
  "decoder_intermediate_size": 4096,
  "decoder_num_attention_heads": 16,
  "decoder_num_hidden_layers": 28,
  "drop_cls_token": true,
  "dtype": "float32",
  "hidden_act": "gelu",
  "hidden_dropout_prob": 0.0,
  "hidden_size": 1152,
  "image_mean": [
    0.5,
    0.5,
    0.5
  ],
  "image_size": 224,
  "image_std": [
    0.5,
    0.5,
    0.5
  ],
  "initializer_range": 0.02,
  "layer_norm_eps": 1e-12,
  "model_type": "f2p_decoder",
  "num_channels": 3,
  "num_patches": 256,
  "patch_size": 14,
  "pretrained_encoder_name": "google/siglip2-so400m-patch14-224",
  "qkv_bias": true,
  "source_decoder_repo": "nyu-visionx/siglip2_decoder",
  "transformers_version": "4.57.6"
}