Factuality Evaluation on Held-out Dataset
8.84Factuality ScoreF-DPO
Evaluation results
| Method | Links | ||
|---|---|---|---|
| F-DPOBackbone=Qwen2.5-14B2026.01 | 8.84 | 0.8 | |
| F-DPOBackbone=Gemma-2-9B2026.01 | 8.26 | 6.8 | |
| Standard DPOBackbone=Gemma-2-9B2026.01 | 8.04 | 9.2 | |
| Base ModelBackbone=Gemma-2-9B2026.01 | 8 | 7.2 | |
| Standard DPOBackbone=Qwen2.5-14B2026.01 | 7.9 | 8 | |
| F-DPOBackbone=Qwen3-8B2026.01 | 7.9 | 8.4 | |
| Base ModelBackbone=Qwen2.5-14B2026.01 | 7.8 | 7.2 | |
| F-DPOBackbone=Qwen2-7B2026.01 | 7.6 | 8.2 | |
| Base ModelBackbone=Gemma-2-2B2026.01 | 7.5 | 9.8 | |
| F-DPOBackbone=Gemma-2-2B2026.01 | 7.3 | 11.6 | |
| Standard DPOBackbone=Gemma-2-2B2026.01 | 7.1 | 14.2 | |
| F-DPOBackbone=LLaMA-3-8B2026.01 | 7 | 15.4 | |
| Base ModelBackbone=Qwen2-7B2026.01 | 6.95 | 18.2 | |
| Standard DPOBackbone=Qwen2-7B2026.01 | 6.5 | 23.8 | |
| Base ModelBackbone=LLaMA-3-8B2026.01 | 6.4 | 25.8 | |
| Standard DPOBackbone=Qwen3-8B2026.01 | 6.14 | 30.2 | |
| Standard DPOBackbone=LLaMA-3-8B2026.01 | 6 | 29 | |
| F-DPOBackbone=Llama-3.2-1B2026.01 | 5.8 | 30 | |
| Base ModelBackbone=Qwen3-8B2026.01 | 5.26 | 42.4 | |
| Standard DPOBackbone=Llama-3.2-1B2026.01 | 5.02 | 40 | |
| Base ModelBackbone=Llama-3.2-1B2026.01 | 5.01 | 43.2 |