Character Reasoning on Letter CD (4,5) (OOD)
14.4AccuracyR1 Distill -> GRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| R1 Distill -> GRPOtraining=SFT + GRPO2025.12 | 14.4 | |
| BOLT -> GRPOtraining=SFT + GRPO2025.12 | 13.1 | |
| SkillFactory -> GRPOtraining=SFT + GRPO2025.12 | 12.8 | |
| RL-Onlytraining=RL2025.12 | 12.5 | |
| Qwen2.5 1.5B Instructtraining=None2025.12 | 10.4 | |
| STaR -> GRPOtraining=SFT + GRPO2025.12 | 9.2 | |
| R1 Distilltraining=SFT2025.12 | 8.8 | |
| SkillFactorytraining=SFT2025.12 | 8.7 | |
| STaRtraining=SFT2025.12 | 7.3 | |
| BOLTtraining=SFT2025.12 | 5.5 |