关联 Issue #41 Add utility to debug reward functions #42 Formal math example: support SFT, data pre-filtering based on correctness, enhance reward fn, tune hyparameters #104 Compile rope while preserving true on policy #3284 Multi-LoRA / Tinker API roadmap #3697 fix: align Qwen3 router precision and P2P staging (after cutoff)
参与讨论