Downloads · 30 days
0
CoF-Reasoning/cof-formal-6k-checkpoints
cof-formal-6k-checkpoints is a machine learning model from CoF-Reasoning. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for pytorch. The card lists the license as apache-2.0.
Private research handoff for the COF formal single-seed program. The source is pinned to mianwu01/COF-maze, branch review/cof-6k-snapshot-20260825, commit 0c7fc59ee3f9ad6fe7f9303141644fee301dbdf4.
Downloads · 30 days
0
Access
Public
Updated Aug 26, 2026
Repo size
229 GB
Likes
0
Public
Click a slice to open those files.
.pt227 GB · 99%
From the Hugging Face model README
Private research handoff for the COF formal single-seed program. The source
is pinned to mianwu01/COF-maze, branch
review/cof-6k-snapshot-20260825, commit
0c7fc59ee3f9ad6fe7f9303141644fee301dbdf4.
For each of five foundation curves, checkpoints/ contains model.pt at 2K,
4K, and 6K. Every model file contains both generator (raw) and
generator_ema. The 6K directories also contain full trainer.pt state for
exact continuation. configs/ contains the corresponding resolved YAML.
results/ contains only dev outputs for the completed bidirectional Maze and
VBVR 2K/4K/6K raw/EMA evaluation. Maze 6K EMA is incomplete (missing ordinal
0031); it is preserved as partial evidence and must not pass the completion
gate. The frozen VBVR decision in selection/ selects 6K raw.
See release_manifest.json for file sizes and SHA256 values. The release has
not read or included sealed final-test data. Base Wan and upstream
Causal-Forcing weights are intentionally referenced as public dependencies in
the manifest instead of duplicated here.
Access is restricted to project collaborators. Do not make this repository public or use dev results to modify the sealed-test protocol.