{ "architectures": [ "LlavaNextForConditionalGeneration" ], "model_type": "llava_next", "vocab_size": 128256, "text_config": { "model_type": "llama", "vocab_size": 128256, "hidden_size": 4096, "intermediate_size": 14336, "num_hidden_layers": 32, "num_attention_heads": 32, "num_key_value_heads": 8, "rope_scaling": { "factor": 8.0, "low_freq_factor": 1.0, "high_freq_factor": 4.0, "original_max_position_embeddings": 8192, "rope_type": "llama3" } }, "vision_config": { "model_type": "clip_vision_model", "hidden_size": 1024, "image_size": 336, "patch_size": 14, "num_hidden_layers": 24, "num_attention_heads": 16 } }