optimum-internal-testing/tiny-random-VisionEncoderDecoderModel-donut

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": 64,13    "decoder_attention_heads": 2,14    "decoder_ffn_dim": 128,15    "decoder_layerdrop": 0.0,16    "decoder_layers": 1,17    "dropout": 0.1,18    "encoder_attention_heads": 2,19    "encoder_ffn_dim": 128,20    "encoder_layerdrop": 0.0,21    "encoder_layers": 1,22    "init_std": 0.02,23    "is_decoder": true,24    "is_encoder_decoder": false,25    "max_position_embeddings": 128,26    "model_type": "mbart",27    "num_hidden_layers": 1,28    "scale_embedding": true,29    "use_cache": true,30    "vocab_size": 5753231  },32  "dtype": "float32",33  "encoder": {34    "attention_probs_dropout_prob": 0.0,35    "depths": [36      1,37      1,38      1,39      140    ],41    "drop_path_rate": 0.1,42    "embed_dim": 8,43    "hidden_act": "gelu",44    "hidden_dropout_prob": 0.0,45    "hidden_size": 64,46    "image_size": [47      2560,48      192049    ],50    "initializer_range": 0.02,51    "layer_norm_eps": 1e-05,52    "mlp_ratio": 4.0,53    "model_type": "donut-swin",54    "num_channels": 3,55    "num_heads": [56      1,57      1,58      1,59      160    ],61    "num_layers": 4,62    "patch_size": 4,63    "path_norm": true,64    "qkv_bias": true,65    "use_absolute_embeddings": false,66    "window_size": 1067  },68  "is_encoder_decoder": true,69  "model_type": "vision-encoder-decoder",70  "tie_word_embeddings": false,71  "transformers_version": "4.57.3"72}73