v6: GRPO with --resume-adapter to start from SFT checkpoint 59ed7f3 verified piyush-mk commited on Apr 25
v5: remove 4-bit quant, variable-length traces, submit oversampling, GRPO exploration fixes 1d75102 verified piyush-mk commited on Apr 25
Fix Qwen3 thinking mode + increase max_new_tokens: training/train_grpo.py 9d40fed verified piyush-mk commited on Apr 25