Text Generation on TruthfulQA
15.5BLEU-4Factual SFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Factual SFTBackbone=Qwen2.5-14B2026.01 | 15.5 | 38.3 | |
| SFTBackbone=Qwen2.5-14B2026.01 | 14.2 | 36.3 | |
| Factual SFT + Standard DPOBackbone=Qwen2.5-14B2026.01 | 12.4 | 34.4 | |
| Base ModelBackbone=Qwen2.5-14B2026.01 | 10.6 | 31.5 | |
| Standard DPOBackbone=Qwen2.5-14B2026.01 | 10.5 | 31.8 | |
| Factual SFT + F-DPOBackbone=Qwen2.5-14B2026.01 | 10.2 | 31.8 | |
| F-DPOBackbone=Qwen2.5-14B2026.01 | 9.9 | 30.6 |