ChristianZ97
/

satp-policy-v2

Reinforcement Learning

theorem-proving

Model card Files Files and versions

2.22 GB

Ctrl+K

Ctrl+K

3 contributors

History: 24 commits

ChristianZ97's picture

card: pin training data (NuminaMath-LEAN via SATP-cleaned@3e78c837)

4fbb2af verified 15 days ago

cache
satp-v2 epoch-7 (val 37.30% / 91-244) + standalone infer.py + premise cache about 1 month ago
.gitattributes

1.58 kB
satp-v2 epoch-7 (val 37.30% / 91-244) + standalone infer.py + premise cache about 1 month ago
README.md

7.38 kB
card: pin training data (NuminaMath-LEAN via SATP-cleaned@3e78c837) 15 days ago
best_checkpoint.pt
Detected Pickle imports (8)
- "torch.ByteStorage",
- "collections.OrderedDict",
- "_codecs.encode",
- "torch._utils._rebuild_tensor_v2",
- "torch.FloatStorage",
- "numpy._core.multiarray._reconstruct",
- "numpy.ndarray",
- "numpy.dtype"
How to fix it?
1.12 GB
xet

satp-v2 epoch-7 (val 37.30% / 91-244) + standalone infer.py + premise cache about 1 month ago
infer.py

33.1 kB
verify_proof: fail closed on degenerate Kimina responses (missing results/response, Error-shaped response) + normalize nullable messages/sorries; name fallback unconditional (codex-review hardening) 29 days ago
reproduce.py

3.98 kB
Add reproduce.py: test-split repro with kimina | lake verifier about 1 month ago