tz_openr1

---

PJ_0509

Check whether the GRPO really improves scores further using for SFT fine-tuned model.

• Model: Qwen2.5-1.5B
• Number Data: 1000(sft),1000(rl),1000(eval)
• Epochs: 20(sft),1by1(rl)
• Methods: SFT,GRPO

---

PJ_0426

Check whether the RLs really improves scores further using for SFT fine-tuned model.

• Model: Qwen2.5-0.5B
• Number Data: 400(sft,rl),400(eval)
• Epochs: 10(sft),1(rl)
• Methods: SFT,GRPO,CPPO,DRGRPO,DRGRPOCPPO,RAFT,REINFORCE

---