divyjx/VideoLLaVA

1{2  "X": [3    "Video",4    "Image"5  ],6  "_name_or_path": "lmsys/vicuna-7b-v1.5",7  "architectures": [8    "LlavaLlamaForCausalLM"9  ],10  "bos_token_id": 1,11  "eos_token_id": 2,12  "freeze_mm_mlp_adapter": false,13  "hidden_act": "silu",14  "hidden_size": 4096,15  "image_aspect_ratio": "pad",16  "image_grid_pinpoints": null,17  "initializer_range": 0.02,18  "intermediate_size": 11008,19  "max_position_embeddings": 4096,20  "mm_hidden_size": 1024,21  "mm_image_tower": "LanguageBind/LanguageBind_Image",22  "mm_projector_type": "mlp2x_gelu",23  "mm_use_x_patch_token": false,24  "mm_use_x_start_end": false,25  "mm_video_tower": "LanguageBind/LanguageBind_Video_merge",26  "mm_vision_select_feature": "patch",27  "mm_vision_select_layer": -2,28  "model_type": "llava",29  "num_attention_heads": 32,30  "num_hidden_layers": 32,31  "num_key_value_heads": 32,32  "pad_token_id": 0,33  "pretraining_tp": 1,34  "rms_norm_eps": 1e-05,35  "rope_scaling": null,36  "tie_word_embeddings": false,37  "torch_dtype": "bfloat16",38  "transformers_version": "4.31.0",39  "tune_mm_mlp_adapter": false,40  "use_cache": true,41  "use_mm_proj": true,42  "vocab_size": 3200043}44