Commonsense Question Answering on CSQA (OOD)
63.8AccuracyR1 Distill -> GRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| R1 Distill -> GRPOtraining=SFT + GRPO2025.12 | 63.8 | |
| BOLT -> GRPOtraining=SFT + GRPO2025.12 | 62.8 | |
| RL-Onlytraining=RL2025.12 | 62.6 | |
| SkillFactory -> GRPOtraining=SFT + GRPO2025.12 | 60.8 | |
| STaR -> GRPOtraining=SFT + GRPO2025.12 | 60.5 | |
| R1 Distilltraining=SFT2025.12 | 56.6 | |
| Qwen2.5 1.5B Instructtraining=None2025.12 | 55.7 | |
| STaRtraining=SFT2025.12 | 55.4 | |
| SkillFactorytraining=SFT2025.12 | 47.1 | |
| BOLTtraining=SFT2025.12 | 46.7 |