trl-internal-testing/tiny-LlavaForConditionalGeneration
0148k
1{2 "architectures": [3 "LlavaForConditionalGeneration"4 ],5 "dtype": "float16",6 "ignore_index": -100,7 "image_seq_length": 576,8 "image_token_index": 32000,9 "model_type": "llava",10 "multimodal_projector_bias": true,11 "pad_token_id": 32001,12 "projector_hidden_act": "gelu",13 "text_config": {14 "_name_or_path": "lmsys/vicuna-7b-v1.5",15 "architectures": [16 "LlamaForCausalLM"17 ],18 "attention_bias": false,19 "attention_dropout": 0.0,20 "dtype": "float16",21 "head_dim": 128,22 "hidden_act": "silu",23 "hidden_size": 16,24 "initializer_range": 0.02,25 "intermediate_size": 11008,26 "layer_types": null,27 "max_position_embeddings": 4096,28 "mlp_bias": false,29 "model_type": "llama",30 "num_attention_heads": 4,31 "num_hidden_layers": 2,32 "num_key_value_heads": 2,33 "pretraining_tp": 1,34 "rms_norm_eps": 1e-05,35 "rope_scaling": null,36 "rope_theta": 10000.0,37 "use_cache": true,38 "vocab_size": 3206439 },40 "tie_word_embeddings": false,41 "transformers_version": "4.56.2",42 "vision_config": {43 "attention_dropout": 0.0,44 "hidden_act": "quick_gelu",45 "hidden_size": 16,46 "image_size": 336,47 "initializer_factor": 1.0,48 "initializer_range": 0.02,49 "intermediate_size": 4096,50 "layer_norm_eps": 1e-05,51 "model_type": "clip_vision_model",52 "num_attention_heads": 4,53 "num_channels": 3,54 "num_hidden_layers": 2,55 "patch_size": 14,56 "projection_dim": 768,57 "vocab_size": 3200058 },59 "vision_feature_layer": -2,60 "vision_feature_select_strategy": "default",61 "vocab_size": 3206462}63 