Science Question Answering on GPQA D
67.4Accuracy (Avg@8)CERL full
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CERL fullTraining=SFT+RL2026.05 | 67.4 | — | |
| HaloTraining=SFT+RL2026.05 | 64.6 | — | |
| Correctness-only RLTraining=RL2026.05 | 64.6 | — | |
| Long-answer SFTTraining=SFT2026.05 | 62.8 | — | |
| Fixed-ratio erasure + free answerTraining=None2026.05 | 62.6 | — | |
| Base Qwen3-8BTraining=N/A2026.05 | 61.9 | — | |
| Length-penalty RLTraining=RL2026.05 | 61.3 | — | |
| TokenSkipTraining=SFT2026.05 | 55.5 | — | |
| IOAStudent Model=Qwen2.5-7B, Teacher LLM=DeepSeek-R12026.02 | — | 16.23 |