Hallucination Mitigation on RAGTruth
98.4FaithfulnessBALTO
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| BALTOModel=Qwen3-8B2026.06 | 98.4 | 71.1 | 70 | — | |
| BALTOModel=Qwen3-4B2026.06 | 94 | 56.7 | 53.3 | — | |
| GRPODModel=Qwen3-4B2026.06 | 93.1 | 50.9 | 47.4 | 35.2 | |
| GRPODModel=Qwen3-8B2026.06 | 92 | 73.8 | 67.9 | 37.2 | |
| GRPOBModel=Qwen3-8B2026.06 | 89.8 | 55.6 | 49.9 | 45.1 | |
| GRPOBModel=Qwen3-4B2026.06 | 89.8 | 21.3 | 19.1 | 39.8 | |
| FSPOModel=Qwen3-8B2026.06 | 73.1 | 84.9 | 62.1 | 34.2 | |
| FSPOModel=Qwen3-4B2026.06 | 73.1 | 84.9 | 62.1 | 34.2 | |
| BaseModel=Qwen3-4B2026.06 | 62.2 | — | 62.2 | — | |
| SFTModel=Qwen3-8B2026.06 | 56.2 | 66.7 | 37.5 | — | |
| DPOModel=Qwen3-8B2026.06 | 52.4 | 11.6 | 6.1 | — | |
| BaseModel=Qwen3-8B2026.06 | 40.4 | — | 40.4 | — |