2 days ago
-
v4: +322 Beam traces (hard tier v2 at levels 5-7, partial 32k rescue), 223 correctness pairs, re-audited (0 key mismatches, 0 leaks)
-
Add TRL-ready training columns and card fixes
- sft_verified: add messages_with_reasoning (<think> reasoning + answer)
- preference_pairs: add TRL-ready chosen/rejected columns
- full_traces: drop reasoning_tokens (endpoint never reported it; always null)
- README: TRL quickstart, provider precision caveat, SFT effort skew, superseded records, distractor note
- package.py updated to produce the same files -
Add check_release.py, reward_fn.py (TRL GRPO), requirements.txt and reproduce section
MarshootKimi
-
Initial release: Beam Verified Reasoning Traces by Marshoot Lab
MarshootKimi
-
initial commit
MarshootKimi