Question Answering on Average (SQuAD, NewsQA, TriviaQA, NQ, HotpotQA)
0.9198FaithFaithRL
Evaluation results
| Method | Links | ||
|---|---|---|---|
| FaithRLModel=Qwen3-1.7B2026.02 | 0.9198 | 0.7355 | |
| KDModel=Qwen3-1.7B2026.02 | 0.9125 | 0.6686 | |
| SFTModel=Qwen3-1.7B2026.02 | 0.9105 | 0.6898 | |
| Self-RefineModel=Qwen3-1.7B2026.02 | 0.9075 | 0.7097 | |
| FSPOModel=Qwen3-1.7B2026.02 | 0.9058 | 0.688 | |
| BaseModel=Qwen3-1.7B2026.02 | 0.9012 | 0.6925 | |
| GRPOModel=Qwen3-1.7B2026.02 | 0.8817 | 0.7119 | |
| FaithRLModel=Qwen3-0.6B2026.02 | 0.844 | 0.6408 | |
| SFTModel=Qwen3-0.6B2026.02 | 0.8328 | 0.6132 | |
| FSPOModel=Qwen3-0.6B2026.02 | 0.8206 | 0.6138 | |
| KDModel=Qwen3-0.6B2026.02 | 0.8159 | 0.6106 | |
| Self-RefineModel=Qwen3-0.6B2026.02 | 0.8156 | 0.6184 | |
| BaseModel=Qwen3-0.6B2026.02 | 0.8132 | 0.6021 | |
| FaithRLModel=DPSK-1.5B2026.02 | 0.8101 | 0.5681 | |
| GRPOModel=Qwen3-0.6B2026.02 | 0.7939 | 0.3859 | |
| GRPOModel=DPSK-1.5B2026.02 | 0.748 | 0.3887 | |
| KDModel=DPSK-1.5B2026.02 | 0.7306 | 0.5193 | |
| SFTModel=DPSK-1.5B2026.02 | 0.697 | 0.4727 | |
| FSPOModel=DPSK-1.5B2026.02 | 0.6936 | 0.5429 | |
| Self-RefineModel=DPSK-1.5B2026.02 | 0.676 | 0.4464 | |
| BaseModel=DPSK-1.5B2026.02 | 0.662 | 0.4487 |