{ "_name_or_path": "google/vivit-b-16x2", "architectures": [ "VivitForVideoClassification" ], "attention_probs_dropout_prob": 0.0, "hidden_act": "gelu_fast", "hidden_dropout_prob": 0.0, "hidden_size": 768, "id2label": { "0": "finish", "1": "fish", "2": "forget", "3": "france", "4": "friend", "5": "girl", "6": "good", "7": "grandfather", "8": "grandmother", "9": "green" }, "image_size": 224, "initializer_range": 0.02, "intermediate_size": 3072, "label2id": { "finish": 0, "fish": 1, "forget": 2, "france": 3, "friend": 4, "girl": 5, "good": 6, "grandfather": 7, "grandmother": 8, "green": 9 }, "layer_norm_eps": 1e-06, "model_type": "vivit", "num_attention_heads": 12, "num_channels": 3, "num_frames": 32, "num_hidden_layers": 12, "qkv_bias": true, "torch_dtype": "float32", "transformers_version": "4.40.0", "tubelet_size": [ 2, 16, 16 ] }