trl-internal-testing/tiny-LlavaNextForConditionalGeneration
0129k
1{2 "architectures": [3 "LlavaNextForConditionalGeneration"4 ],5 "dtype": "bfloat16",6 "ignore_index": -100,7 "image_grid_pinpoints": [8 [9 336,10 67211 ],12 [13 672,14 33615 ],16 [17 672,18 67219 ],20 [21 1008,22 33623 ],24 [25 336,26 100827 ]28 ],29 "image_seq_length": 576,30 "image_token_index": 32000,31 "model_type": "llava_next",32 "multimodal_projector_bias": true,33 "projector_hidden_act": "gelu",34 "text_config": {35 "_name_or_path": "mistralai/Mistral-7B-Instruct-v0.2",36 "architectures": [37 "MistralForCausalLM"38 ],39 "attention_dropout": 0.0,40 "head_dim": null,41 "hidden_act": "silu",42 "hidden_size": 16,43 "initializer_range": 0.02,44 "intermediate_size": 14336,45 "layer_types": null,46 "max_position_embeddings": 32768,47 "model_type": "mistral",48 "num_attention_heads": 4,49 "num_hidden_layers": 2,50 "num_key_value_heads": 2,51 "rms_norm_eps": 1e-05,52 "rope_theta": 1000000.0,53 "sliding_window": null,54 "use_cache": true,55 "vocab_size": 3206456 },57 "tie_word_embeddings": false,58 "transformers_version": "4.56.2",59 "use_image_newline_parameter": true,60 "vision_config": {61 "attention_dropout": 0.0,62 "hidden_act": "quick_gelu",63 "hidden_size": 16,64 "image_size": 336,65 "initializer_factor": 1.0,66 "initializer_range": 0.02,67 "intermediate_size": 4096,68 "layer_norm_eps": 1e-05,69 "model_type": "clip_vision_model",70 "num_attention_heads": 4,71 "num_channels": 3,72 "num_hidden_layers": 2,73 "patch_size": 14,74 "projection_dim": 768,75 "vocab_size": 3200076 },77 "vision_feature_layer": -2,78 "vision_feature_select_strategy": "default",79 "vocab_size": 3206480}81 