Commit History
-
v5d: save best-epoch checkpoint during SFT
Verified signaturea674764 -
v5c: add --min-investigation-steps filter for submit-only with longer contexts
Verified signature2857a20 -
v5b: add --submit-only SFT mode
Verified signature2a667e7 -
v5: remove 4-bit quant, variable-length traces, submit oversampling, GRPO exploration fixes
Verified signature1d75102 -
Upload folder using huggingface_hub
Verified signature4cd2128 -
Fix Qwen3 thinking mode + increase max_new_tokens: training/train_sft.py
Verified signature7c48dcf -
Add InvoiceGuard SFT and merge tooling
Verified signature25ee43d