Short-form Question Answering on 2wiki (test)
26.5EMPPO + PRS_short
Evaluation Results
| Method | Links | |
|---|---|---|
| PPO + PRS_shortBackbone=Qwen2.5-3B-Instruct2025.12 | 26.5 | |
| VSPO + PRS_shortBackbone=Qwen2.5-3B-Instruct2025.12 | 26.5 | |
| GRPO + PRS_shortBackbone=Qwen2.5-3B-Instruct2025.12 | 26 | |
| SFTBackbone=Qwen2.5-3B-Instruct2025.12 | 24.8 | |
| UntrainedBackbone=Qwen2.5-3B-Instruct2025.12 | 8.91 |