V-JEPA 2 ViT-H/16

https://arxiv.org/abs/2506.09985

Lucid port of facebook/vjepa2-vith-fpc64-256/model.safetensors, converted to Lucid-native safetensors.

Available weights

Usage

import lucid.models as models
from lucid.models.weights import Vjepa2VitHugeWeights

# default tag
model = models.vjepa2_vit_huge(pretrained=True)

# explicit tag (enum or string)
model = models.vjepa2_vit_huge(weights=Vjepa2VitHugeWeights.FPC64_256)
model = models.vjepa2_vit_huge(pretrained="FPC64_256")

# preprocessing travels with the weights
weights = Vjepa2VitHugeWeights.FPC64_256
preprocess = weights.transforms()
out = model(preprocess(image)[None])
logits = out.logits  # (B, num_classes)

Conversion

Converted from facebook/vjepa2-vith-fpc64-256/model.safetensors via python -m tools.convert_weights vjepa2_vit_huge --tag FPC64_256. Key mapping + numerical parity verified against the source.

License

apache-2.0 — inherited from the original weights.

Citation

Assran et al., "V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning," arXiv:2506.09985, 2025.