Science Question Answering on MMLU-Pro OOD
54.7Accuracy (MMLU-Pro OOD Science)RL-PLUS
Evaluation results
| Method | Links | |
|---|---|---|
| RL-PLUSBase model=Qwen2.5-Math-7B2025.07 | 54.7 | |
| GRPOBase model=Qwen2.5-Math-7B2025.07 | 47.5 | |
| SFTBase model=Qwen2.5-Math-7B2025.07 | 42.7 | |
| SFT+GRPOBase model=Qwen2.5-Math-7B2025.07 | 37.7 | |
| Base ModelBase model=Qwen2.5-Math-7B2025.07 | 30.2 |