{ "architectures": [ "GroupViTModel" ], "dtype": "float32", "initializer_factor": 1.0, "initializer_range": 0.02, "logit_scale_init_value": 2.6592, "model_type": "groupvit", "output_segmentation": false, "projection_dim": 64, "projection_intermediate_dim": 4096, "text_config": { "attention_dropout": 0.1, "bos_token_id": 49406, "dropout": 0.1, "eos_token_id": 49407, "hidden_act": "quick_gelu", "hidden_size": 32, "initializer_factor": 1.0, "initializer_range": 0.02, "intermediate_size": 37, "layer_norm_eps": 1e-05, "max_position_embeddings": 512, "model_type": "groupvit_text_model", "num_attention_heads": 4, "num_hidden_layers": 2, "pad_token_id": 49407, "vocab_size": 49408 }, "transformers_version": "5.16.0.dev0", "vision_config": { "assign_eps": 1.0, "assign_mlp_ratio": [ 0.5, 4 ], "attention_dropout": 0.1, "depths": [ 6, 3, 3 ], "dropout": 0.1, "hidden_act": "gelu", "hidden_size": 32, "image_size": 30, "initializer_factor": 1.0, "initializer_range": 0.02, "intermediate_size": 37, "layer_norm_eps": 1e-05, "model_type": "groupvit_vision_model", "num_attention_heads": 4, "num_channels": 3, "num_group_tokens": [ 64, 8, 0 ], "num_hidden_layers": 12, "num_output_groups": [ 64, 8, 8 ], "patch_size": 2 } }