Question Answering on TriviaQA (Faith and Accuracy)
87.48FaithFaithRL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FaithRLModel=Qwen3-1.7B2026.02 | 87.48 | 78.4 | |
| SFTModel=Qwen3-1.7B2026.02 | 87.1 | 76.26 | |
| KDModel=Qwen3-1.7B2026.02 | 86.84 | 75.36 | |
| FSPOModel=Qwen3-1.7B2026.02 | 85.93 | 76.47 | |
| GRPOModel=Qwen3-1.7B2026.02 | 85.88 | 71.74 | |
| Self-RefineModel=Qwen3-1.7B2026.02 | 85.61 | 77.59 | |
| BaseModel=Qwen3-1.7B2026.02 | 85.4 | 78.48 | |
| FaithRLModel=Qwen3-0.6B2026.02 | 79.63 | 69.05 | |
| KDModel=Qwen3-0.6B2026.02 | 77.8 | 67.14 | |
| FSPOModel=Qwen3-0.6B2026.02 | 77.64 | 66.82 | |
| BaseModel=Qwen3-0.6B2026.02 | 77.4 | 67.99 | |
| Self-RefineModel=Qwen3-0.6B2026.02 | 76.57 | 68.42 | |
| SFTModel=Qwen3-0.6B2026.02 | 76.42 | 66.57 | |
| FaithRLModel=DPSK-1.5B2026.02 | 76.2 | 55.03 | |
| GRPOModel=Qwen3-0.6B2026.02 | 74.61 | 35.78 | |
| KDModel=DPSK-1.5B2026.02 | 69.19 | 53.7 | |
| FSPOModel=DPSK-1.5B2026.02 | 68.23 | 60.84 | |
| GRPOModel=DPSK-1.5B2026.02 | 66.81 | 37.24 | |
| SFTModel=DPSK-1.5B2026.02 | 66.28 | 54.98 | |
| Self-RefineModel=DPSK-1.5B2026.02 | 65.2 | 54.03 | |
| BaseModel=DPSK-1.5B2026.02 | 61 | 51.46 |