Hallucination Mitigation on ConFiQA
99FaithBALTO
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| BALTOModel=Qwen3-4B2026.06 | 99 | 96.2 | 95.2 | 38.6 | |
| GRPOBModel=Qwen3-4B2026.06 | 97.8 | 96.3 | 94.2 | 36.2 | |
| BALTOModel=Qwen3-8B2026.06 | 97.5 | 96.2 | 93.8 | 80.2 | |
| FSPOModel=Qwen3-8B2026.06 | 94.8 | 97.3 | 92.2 | 66.4 | |
| FSPOModel=Qwen3-4B2026.06 | 94.8 | 97.3 | 92.2 | 66.4 | |
| GRPODModel=Qwen3-8B2026.06 | 94.5 | 95.2 | 90 | 74.4 | |
| GRPOBModel=Qwen3-8B2026.06 | 92.8 | 93.5 | 86.8 | 73.8 | |
| GRPODModel=Qwen3-4B2026.06 | 88 | 92.8 | 81.7 | 37.8 | |
| BaseModel=Qwen3-4B2026.06 | 80.3 | — | 80.3 | 36.8 | |
| DPOModel=Qwen3-8B2026.06 | 72.8 | 84 | 61.2 | 62 | |
| SFTModel=Qwen3-8B2026.06 | 71.2 | 68 | 48.4 | 71.6 | |
| BaseModel=Qwen3-8B2026.06 | 68.2 | — | 68.2 | 70.6 |