| { | |
| "model_type": "llava", | |
| "architectures": [ | |
| "PrefixConditioningVLM" | |
| ], | |
| "vision_config": { | |
| "model_type": "siglip_vision_model", | |
| "hidden_size": 768, | |
| "image_size": 224, | |
| "patch_size": 16 | |
| }, | |
| "text_config": { | |
| "model_type": "gpt2", | |
| "vocab_size": 49152, | |
| "n_positions": 2048, | |
| "n_embd": 576, | |
| "n_layer": 30, | |
| "n_head": 9 | |
| }, | |
| "projector_hidden_size": 512, | |
| "vision_feature_select_strategy": "full", | |
| "vision_feature_layer": -1 | |
| } |