Question Answering on NQ (Faith, Accuracy)
0.9083FaithFaithRL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FaithRLModel=Qwen3-1.7B2026.02 | 0.9083 | 0.7305 | |
| Self-RefineModel=Qwen3-1.7B2026.02 | 0.9062 | 0.6846 | |
| SFTModel=Qwen3-1.7B2026.02 | 0.9043 | 0.6902 | |
| KDModel=Qwen3-1.7B2026.02 | 0.9035 | 0.6393 | |
| FSPOModel=Qwen3-1.7B2026.02 | 0.9022 | 0.6727 | |
| GRPOModel=Qwen3-1.7B2026.02 | 0.898 | 0.7139 | |
| BaseModel=Qwen3-1.7B2026.02 | 0.894 | 0.6795 | |
| FaithRLModel=Qwen3-0.6B2026.02 | 0.8044 | 0.6531 | |
| SFTModel=Qwen3-0.6B2026.02 | 0.798 | 0.6234 | |
| FSPOModel=Qwen3-0.6B2026.02 | 0.7942 | 0.6239 | |
| GRPOModel=Qwen3-0.6B2026.02 | 0.7921 | 0.3976 | |
| KDModel=Qwen3-0.6B2026.02 | 0.7886 | 0.614 | |
| Self-RefineModel=Qwen3-0.6B2026.02 | 0.7883 | 0.6382 | |
| BaseModel=Qwen3-0.6B2026.02 | 0.788 | 0.6102 | |
| FaithRLModel=DPSK-1.5B2026.02 | 0.786 | 0.5938 | |
| KDModel=DPSK-1.5B2026.02 | 0.7148 | 0.5325 | |
| GRPOModel=DPSK-1.5B2026.02 | 0.7124 | 0.4276 | |
| FSPOModel=DPSK-1.5B2026.02 | 0.6862 | 0.5685 | |
| SFTModel=DPSK-1.5B2026.02 | 0.6805 | 0.4911 | |
| BaseModel=DPSK-1.5B2026.02 | 0.626 | 0.4608 | |
| Self-RefineModel=DPSK-1.5B2026.02 | 0.6229 | 0.459 |