config.json
| 1 | { |
| 2 | "architectures": [ |
| 3 | "VJEPA2Model" |
| 4 | ], |
| 5 | "attention_probs_dropout_prob": 0.0, |
| 6 | "crop_size": 256, |
| 7 | "drop_path_rate": 0.0, |
| 8 | "frames_per_clip": 64, |
| 9 | "hidden_act": "gelu", |
| 10 | "hidden_dropout_prob": 0.0, |
| 11 | "hidden_size": 1024, |
| 12 | "image_size": 256, |
| 13 | "in_chans": 3, |
| 14 | "initializer_range": 0.02, |
| 15 | "layer_norm_eps": 1e-06, |
| 16 | "mlp_ratio": 4, |
| 17 | "model_type": "vjepa2", |
| 18 | "num_attention_heads": 16, |
| 19 | "num_hidden_layers": 24, |
| 20 | "patch_size": 16, |
| 21 | "pred_hidden_size": 384, |
| 22 | "pred_mlp_ratio": 4.0, |
| 23 | "pred_num_attention_heads": 12, |
| 24 | "pred_num_hidden_layers": 12, |
| 25 | "pred_num_mask_tokens": 10, |
| 26 | "pred_zero_init_mask_tokens": true, |
| 27 | "qkv_bias": true, |
| 28 | "torch_dtype": "float32", |
| 29 | "transformers_version": "4.53.0.dev0", |
| 30 | "tubelet_size": 2, |
| 31 | "use_SiLU": false, |
| 32 | "wide_SiLU": true |
| 33 | } |
| 34 | |