{ "architectures": [ "VJEPA2Model" ], "attention_dropout": 0.0, "attention_probs_dropout_prob": 0.0, "crop_size": 16, "drop_path_rate": 0.0, "dtype": "float32", "frames_per_clip": 2, "hidden_act": "gelu", "hidden_size": 32, "in_chans": 3, "initializer_range": 0.02, "layer_norm_eps": 1e-06, "mlp_ratio": 1.0, "model_type": "vjepa2", "num_attention_heads": 2, "num_hidden_layers": 2, "num_pooler_layers": 3, "patch_size": 16, "pred_hidden_size": 32, "pred_mlp_ratio": 4.0, "pred_num_attention_heads": 2, "pred_num_hidden_layers": 2, "pred_num_mask_tokens": 10, "pred_zero_init_mask_tokens": true, "qkv_bias": true, "transformers_version": "5.16.0.dev0", "tubelet_size": 2 }