_target_: models.encoder.vjepa2.VJEPA2Encoder
name: "vjepa2"
feature_key: "x_norm_patchtokens"
output_dim: 1024
n_patches: 256