Question Answering on NQ-Open In-Domain (test)
58.13PrecisionFAITH
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FAITHBackbone=Llama3-8B2026.04 | 58.13 | 27.58 | |
| FAITH_sft+ppoBackbone=Llama3-8B2026.04 | 57.99 | 26.52 | |
| ICL-CoTBackbone=Llama3-8B2026.04 | 57.34 | 23.6 | |
| InFactBackbone=Llama3-8B2026.04 | 56.93 | 26.11 | |
| Context-DPOBackbone=Llama3-8B2026.04 | 56.82 | 26.02 | |
| UAlign_ragBackbone=Llama3-8B2026.04 | 56.78 | 26.1 | |
| DTABackbone=Llama3-8B2026.04 | 56.72 | 21.12 | |
| FAITH_sftBackbone=Llama3-8B2026.04 | 56.72 | 26.21 | |
| UAlignBackbone=Llama3-8B2026.04 | 56.6 | 26.09 | |
| UAlign_sftBackbone=Llama3-8B2026.04 | 56.02 | 25.49 | |
| ICL-CoTBackbone=Mistral-7B-v0.12026.04 | 54.47 | 18.22 | |
| UAlign_ragBackbone=Mistral-7B-v0.12026.04 | 54.47 | 20.01 | |
| UAlignBackbone=Mistral-7B-v0.12026.04 | 54.17 | 19.64 | |
| SFT_ragBackbone=Mistral-7B-v0.12026.04 | 51.03 | 14.53 | |
| RL-DPOBackbone=Mistral-7B-v0.12026.04 | 50.95 | 16.42 | |
| SFTBackbone=Mistral-7B-v0.12026.04 | 50.82 | 14.42 | |
| RL-DPOBackbone=Llama3-8B2026.04 | 49.65 | 19.18 | |
| FAITHBackbone=Mistral-7B-v0.12026.04 | 48.05 | 23.91 | |
| Context-DPOBackbone=Mistral-7B-v0.12026.04 | 47.21 | 22.51 | |
| InFactBackbone=Mistral-7B-v0.12026.04 | 47.01 | 22.33 | |
| FAITH_sft+ppoBackbone=Mistral-7B-v0.12026.04 | 46.6 | 23.19 | |
| FAITH_sftBackbone=Mistral-7B-v0.12026.04 | 46.16 | 22.96 | |
| UAlign_sftBackbone=Mistral-7B-v0.12026.04 | 43.32 | 21.55 | |
| SFT_ragBackbone=Llama3-8B2026.04 | 42.64 | 17.58 | |
| SFTBackbone=Llama3-8B2026.04 | 41.41 | 16.57 | |
| DTABackbone=Mistral-7B-v0.12026.04 | 41.01 | 20.49 |