WARP-RM Sim Bottles SSS15
Paper reward-model checkpoint for the simulated bottles task, trained with stride-15 sequence scoring. The checkpoint contains the WARP-RM aggregation head only; it does not include DINOv3 weights.
Use it with the public WARP-RM release code and
uynitsuj/sim-bottles-mjwarp-v1. Obtain the matching DINOv3 backbone
separately from Meta/Hugging Face and comply with its license terms; this
repository does not redistribute DINOv3 weights.
Files and SHA256 checksums are listed in MANIFEST.json.