| { | |
| "architectures": [ | |
| "LlavaForConditionalGeneration" | |
| ], | |
| "ignore_index": -100, | |
| "image_seq_length": 1, | |
| "image_token_index": 10, | |
| "model_type": "llava", | |
| "projector_hidden_act": "gelu", | |
| "text_config": { | |
| "hidden_size": 12288, | |
| "intermediate_size": 28672, | |
| "max_position_embeddings": 131072, | |
| "model_type": "mistral", | |
| "norm_eps": 1e-05, | |
| "num_attention_heads": 96, | |
| "num_hidden_layers": 88, | |
| "num_key_value_heads": 8, | |
| "rope_theta": 1000000000.0, | |
| "sliding_window": 4096, | |
| "vocab_size": 32768 | |
| }, | |
| "torch_dtype": "bfloat16", | |
| "transformers_version": "4.47.0.dev0", | |
| "vision_config": { | |
| "head_dim": 88, | |
| "hidden_act": "silu", | |
| "hidden_size": 1408, | |
| "image_size": 1024, | |
| "image_token_id": 10, | |
| "intermediate_size": 6144, | |
| "model_type": "pixtral", | |
| "num_hidden_layers": 40, | |
| "patch_size": 16, | |
| "rope_theta": 10000.0 | |
| }, | |
| "vision_feature_layer": -1, | |
| "vision_feature_select_strategy": "full" | |
| } | |