Post-RL Evaluation on BiGGen Bench
63.7AccuracyMRRG
Evaluation Results
| Method | Links | |
|---|---|---|
| MRRG2026.07 | 63.7 | |
| SVRGSub-reward (SR)=true2026.07 | 62 | |
| Chasing the Tail2026.07 | 61.1 | |
| SVRGSub-reward (SR)=false2026.07 | 60 | |
| Base policyBackbone=Qwen2.5-3B-Instruct2026.07 | 57.8 |