| { | |
| "additional_vocab_size": 40, | |
| "architectures": [ | |
| "BiModernVBert" | |
| ], | |
| "auto_map": { | |
| "AutoConfig": "configuration_vbert.VBertConfig", | |
| "AutoModel": "modeling_vbert.VBertModel", | |
| "AutoModelForMaskedLM": "modeling_vbert.VBertForMaskedLM" | |
| }, | |
| "freeze_config": { | |
| "freeze_lm_head": true, | |
| "freeze_text_layers": true, | |
| "freeze_vision_layers": true | |
| }, | |
| "hidden_size": 768, | |
| "image_token_id": 50407, | |
| "initializer_range": 0.02, | |
| "max_position_embeddings": 8192, | |
| "model_type": "vbert", | |
| "neftune_noise_alpha": 0.0, | |
| "output_attentions": false, | |
| "pixel_shuffle_factor": 4, | |
| "qk_layer_norms": false, | |
| "scale_factor": 4, | |
| "text_config": { | |
| "hidden_size": 768, | |
| "intermediate_size": 1152, | |
| "mlp_bias": false, | |
| "model_type": "vbert", | |
| "num_hidden_layers": 22, | |
| "text_model_name": "jhu-clsp/ettin-encoder-150m", | |
| "vocab_size": 50368 | |
| }, | |
| "tie_word_embeddings": false, | |
| "torch_dtype": "float32", | |
| "transformers_version": null, | |
| "use_cache": true, | |
| "use_resampler": false, | |
| "vision_config": { | |
| "embed_dim": 768, | |
| "image_size": 512, | |
| "intermediate_size": 3072, | |
| "model_type": "vbert", | |
| "num_hidden_layers": 12, | |
| "patch_size": 16, | |
| "vision_model_name": "google/siglip2-base-patch16-512" | |
| }, | |
| "vocab_size": 50368 | |
| } | |