subhodipbanerjee/kyc_classification

1{2  "architectures": [3    "VisionEncoderDecoderModel"4  ],5  "decoder": {6    "activation_dropout": 0.0,7    "activation_function": "gelu",8    "add_cross_attention": true,9    "add_final_layer_norm": true,10    "attention_dropout": 0.0,11    "classifier_dropout": 0.0,12    "d_model": 1024,13    "decoder_attention_heads": 16,14    "decoder_ffn_dim": 4096,15    "decoder_layerdrop": 0.0,16    "decoder_layers": 4,17    "dropout": 0.1,18    "encoder_attention_heads": 16,19    "encoder_ffn_dim": 4096,20    "encoder_layerdrop": 0.0,21    "encoder_layers": 12,22    "init_std": 0.02,23    "is_decoder": true,24    "is_encoder_decoder": false,25    "max_length": 768,26    "max_position_embeddings": 1536,27    "model_type": "mbart",28    "num_hidden_layers": 12,29    "scale_embedding": true,30    "torch_dtype": "float32",31    "use_cache": true,32    "vocab_size": 5752533  },34  "encoder": {35    "attention_probs_dropout_prob": 0.0,36    "depths": [37      2,38      2,39      14,40      241    ],42    "drop_path_rate": 0.1,43    "embed_dim": 128,44    "hidden_act": "gelu",45    "hidden_dropout_prob": 0.0,46    "hidden_size": 1024,47    "image_size": [48      1280,49      96050    ],51    "initializer_range": 0.02,52    "layer_norm_eps": 1e-05,53    "mlp_ratio": 4.0,54    "model_type": "donut-swin",55    "num_channels": 3,56    "num_heads": [57      4,58      8,59      16,60      3261    ],62    "num_layers": 4,63    "patch_size": 4,64    "path_norm": true,65    "qkv_bias": true,66    "torch_dtype": "float32",67    "use_absolute_embeddings": false,68    "window_size": 1069  },70  "is_encoder_decoder": true,71  "model_type": "vision-encoder-decoder",72  "tie_word_embeddings": false,73  "torch_dtype": "float32",74  "transformers_version": "4.52.4"75}76