| { |
| "architectures": [ |
| "Pix2StructForConditionalGeneration" |
| ], |
| "decoder_start_token_id": 0, |
| "dtype": "float32", |
| "eos_token_id": 1, |
| "initializer_factor": 1.0, |
| "initializer_range": 0.02, |
| "is_encoder_decoder": true, |
| "is_vqa": false, |
| "model_type": "pix2struct", |
| "pad_token_id": 0, |
| "text_config": { |
| "add_cross_attention": false, |
| "bos_token_id": null, |
| "cross_attention_hidden_size": null, |
| "d_ff": 2048, |
| "d_kv": 64, |
| "decoder_start_token_id": 0, |
| "dense_act_fn": "gelu_new", |
| "dropout_rate": 0.05, |
| "dtype": "float32", |
| "encoder_hidden_size": 768, |
| "eos_token_id": 1, |
| "feed_forward_proj": "gated-gelu", |
| "finetuning_task": null, |
| "hidden_size": 768, |
| "initializer_factor": 1.0, |
| "initializer_range": 0.02, |
| "is_decoder": true, |
| "is_encoder_decoder": true, |
| "is_gated_act": true, |
| "layer_norm_epsilon": 1e-06, |
| "model_type": "pix2struct_text_model", |
| "num_heads": 12, |
| "num_layers": 12, |
| "pad_token_id": 0, |
| "prefix": null, |
| "pruned_heads": {}, |
| "relative_attention_max_distance": 128, |
| "relative_attention_num_buckets": 32, |
| "sep_token_id": null, |
| "task_specific_params": null, |
| "tf_legacy_loss": false, |
| "tie_encoder_decoder": false, |
| "tie_word_embeddings": false, |
| "tokenizer_class": null, |
| "torchscript": false, |
| "use_bfloat16": false, |
| "use_cache": false, |
| "vocab_size": 50432 |
| }, |
| "tie_word_embeddings": false, |
| "transformers_version": "5.0.0", |
| "use_cache": false, |
| "vision_config": { |
| "add_cross_attention": false, |
| "attention_dropout": 0.05, |
| "bos_token_id": null, |
| "cross_attention_hidden_size": null, |
| "d_ff": 2048, |
| "d_kv": 64, |
| "decoder_start_token_id": null, |
| "dense_act_fn": "gelu_new", |
| "dropout_rate": 0.06, |
| "dtype": "float32", |
| "eos_token_id": null, |
| "finetuning_task": null, |
| "hidden_dropout_prob": 0.05, |
| "hidden_size": 768, |
| "image_size": 384, |
| "initializer_factor": 1.0, |
| "initializer_range": 0.02, |
| "is_decoder": false, |
| "layer_norm_bias": false, |
| "layer_norm_eps": 1e-06, |
| "mlp_bias": false, |
| "model_type": "pix2struct_vision_model", |
| "num_attention_heads": 12, |
| "num_channels": 3, |
| "num_hidden_layers": 12, |
| "pad_token_id": null, |
| "patch_embed_hidden_size": 768, |
| "patch_size": 16, |
| "prefix": null, |
| "projection_dim": 768, |
| "pruned_heads": {}, |
| "qkv_bias": false, |
| "relative_attention_max_distance": 128, |
| "relative_attention_num_buckets": 32, |
| "sep_token_id": null, |
| "seq_len": 4096, |
| "task_specific_params": null, |
| "tf_legacy_loss": false, |
| "tie_encoder_decoder": false, |
| "tie_word_embeddings": true, |
| "tokenizer_class": null, |
| "torchscript": false, |
| "use_bfloat16": false |
| } |
| } |
|
|
