vjepa_large_16_cls

→VJEPAForVideoClassification

vjepa_large_16_cls(pretrained: bool = False, overrides: object = {})

V-JEPA ViT-L/16 under the paper's attentive probe.

pretrainedbool= False

Not redistributed; True raises.

**overridesobject= {}

Optional VJEPAConfig field overrides.

VJEPAForVideoClassification

The pretraining networks, an attentive pooler and a classifier.

Reference: Bardes et al., arXiv:2404.08471, Section 4.3. num_classes defaults to 400, Kinetics-400's count.

>>> from lucid.models import AutoConfig
>>> AutoConfig.from_pretrained("vjepa_large_16_cls").num_classes
400