trl-internal-testing/tiny-InternVLForConditionalGeneration
051k
1{2 "architectures": [3 "InternVLForConditionalGeneration"4 ],5 "downsample_ratio": 0.5,6 "dtype": "bfloat16",7 "image_seq_length": 256,8 "image_token_id": 151667,9 "model_type": "internvl",10 "projector_hidden_act": "gelu",11 "text_config": {12 "architectures": [13 "Qwen2ForCausalLM"14 ],15 "attention_dropout": 0.0,16 "bos_token_id": 151643,17 "dtype": "bfloat16",18 "eos_token_id": 151645,19 "hidden_act": "silu",20 "hidden_size": 16,21 "initializer_range": 0.02,22 "intermediate_size": 18944,23 "layer_types": [24 "full_attention",25 "full_attention"26 ],27 "max_position_embeddings": 32768,28 "max_window_layers": 70,29 "model_type": "qwen2",30 "num_attention_heads": 4,31 "num_hidden_layers": 2,32 "num_key_value_heads": 2,33 "rms_norm_eps": 1e-06,34 "rope_scaling": {35 "factor": 2.0,36 "rope_type": "dynamic",37 "type": "dynamic"38 },39 "rope_theta": 1000000.0,40 "sliding_window": null,41 "use_cache": true,42 "use_sliding_window": false,43 "vocab_size": 15167444 },45 "transformers_version": "4.57.3",46 "vision_config": {47 "architectures": [48 "InternVisionModel"49 ],50 "attention_bias": true,51 "attention_dropout": 0.0,52 "dropout": 0.0,53 "dtype": "bfloat16",54 "embed_dim": 64,55 "hidden_act": "gelu",56 "hidden_dropout_prob": 0.0,57 "hidden_size": 16,58 "image_size": [59 448,60 44861 ],62 "initializer_factor": 0.1,63 "initializer_range": 1e-10,64 "intermediate_size": 4096,65 "layer_norm_eps": 1e-06,66 "layer_scale_init_value": 0.1,67 "model_type": "internvl_vision",68 "norm_type": "layer_norm",69 "num_attention_heads": 4,70 "num_channels": 3,71 "num_hidden_layers": 2,72 "num_key_value_heads": 2,73 "patch_size": [74 14,75 1476 ],77 "projection_dropout": 0.0,78 "use_absolute_position_embeddings": true,79 "use_mask_token": false,80 "use_mean_pooling": true,81 "use_qk_norm": false82 },83 "vision_feature_layer": -1,84 "vision_feature_select_strategy": "default"85}86 