FPC64 ViT-g/16 representation weights at 256 pixels.
Converted from facebook/vjepa2-vitg-fpc64-256. This is the encoder
V-JEPA 2-AC is post-trained from, and the only one that widens its
feed-forward to 48 / 11. Reproduces its source to 1.0e-4
relative — just over the parity tool's global bound, which was left
where it is because the gap tracks depth rather than a defect.