Question Answering on SciQ In-Domain (test)
83.68PrecisionFAITH_sft+ppo
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FAITH_sft+ppoBackbone=Mistral-7B-v0.12026.04 | 83.68 | 51.8 | |
| FAITH_sftBackbone=Mistral-7B-v0.12026.04 | 82.88 | 51.3 | |
| FAITHBackbone=Mistral-7B-v0.12026.04 | 81.42 | 51.4 | |
| FAITHBackbone=Llama3-8B2026.04 | 80.61 | 49.99 | |
| FAITH_sft+ppoBackbone=Llama3-8B2026.04 | 80.29 | 49.7 | |
| FAITH_sftBackbone=Llama3-8B2026.04 | 78.85 | 48.65 | |
| UAlign_ragBackbone=Llama3-8B2026.04 | 77.47 | 48.25 | |
| Context-DPOBackbone=Llama3-8B2026.04 | 77.32 | 48.62 | |
| InFactBackbone=Llama3-8B2026.04 | 76.94 | 48.21 | |
| UAlignBackbone=Llama3-8B2026.04 | 76.44 | 48 | |
| UAlign_sftBackbone=Llama3-8B2026.04 | 75.87 | 47.65 | |
| InFactBackbone=Mistral-7B-v0.12026.04 | 75.32 | 48.03 | |
| Context-DPOBackbone=Mistral-7B-v0.12026.04 | 75.13 | 47.89 | |
| UAlign_ragBackbone=Mistral-7B-v0.12026.04 | 74.22 | 46.89 | |
| SFT_ragBackbone=Llama3-8B2026.04 | 73.23 | 45.97 | |
| UAlignBackbone=Mistral-7B-v0.12026.04 | 73.21 | 46.7 | |
| UAlign_sftBackbone=Mistral-7B-v0.12026.04 | 72.45 | 45.87 | |
| ICL-CoTBackbone=Llama3-8B2026.04 | 72.34 | 45.9 | |
| SFTBackbone=Llama3-8B2026.04 | 72.18 | 45.4 | |
| ICL-CoTBackbone=Mistral-7B-v0.12026.04 | 71.87 | 44.2 | |
| RL-DPOBackbone=Llama3-8B2026.04 | 71.23 | 44.2 | |
| RL-DPOBackbone=Mistral-7B-v0.12026.04 | 66.44 | 41.8 | |
| SFT_ragBackbone=Mistral-7B-v0.12026.04 | 66.08 | 43.22 | |
| SFTBackbone=Mistral-7B-v0.12026.04 | 65.85 | 42.5 |