Question Answering on HotpotQA (Faith and Accuracy)
93.4FaithfulnessFaithRL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FaithRLModel=Qwen3-1.7B2026.02 | 93.4 | 76.98 | |
| KDModel=Qwen3-1.7B2026.02 | 92.57 | 73.1 | |
| FSPOModel=Qwen3-1.7B2026.02 | 91.83 | 75.46 | |
| SFTModel=Qwen3-1.7B2026.02 | 91.46 | 73.74 | |
| Self-RefineModel=Qwen3-1.7B2026.02 | 90.61 | 72.46 | |
| BaseModel=Qwen3-1.7B2026.02 | 90.4 | 72.81 | |
| FaithRLModel=Qwen3-0.6B2026.02 | 86.4 | 66.46 | |
| GRPOModel=Qwen3-1.7B2026.02 | 85.6 | 74.56 | |
| FaithRLModel=DPSK-1.5B2026.02 | 85.4 | 62.74 | |
| FSPOModel=Qwen3-0.6B2026.02 | 84.61 | 63.87 | |
| KDModel=Qwen3-0.6B2026.02 | 83.64 | 65.42 | |
| SFTModel=Qwen3-0.6B2026.02 | 83.29 | 65.35 | |
| GRPOModel=DPSK-1.5B2026.02 | 81.8 | 40.12 | |
| BaseModel=Qwen3-0.6B2026.02 | 81.6 | 63.21 | |
| Self-RefineModel=Qwen3-0.6B2026.02 | 80.79 | 64.27 | |
| GRPOModel=Qwen3-0.6B2026.02 | 79.85 | 35.6 | |
| KDModel=DPSK-1.5B2026.02 | 76.6 | 57.46 | |
| SFTModel=DPSK-1.5B2026.02 | 68.65 | 52.4 | |
| Self-RefineModel=DPSK-1.5B2026.02 | 68.27 | 48.65 | |
| FSPOModel=DPSK-1.5B2026.02 | 67.65 | 57.3 | |
| BaseModel=DPSK-1.5B2026.02 | 66.6 | 49.2 |