V-JEPA2 — Meta's self-supervised video ViT (ViT-G and friends).
Ships encoder, predictor, attentive pooler, and IR graph paths for all
three when .device(...) is set on the runner.
V-JEPA2 — Meta's self-supervised video ViT (ViT-G and friends).
Ships encoder, predictor, attentive pooler, and IR graph paths for all
three when .device(...) is set on the runner.