Logical Refinement of Natural Language Explanations on QASC
17Initial ScoreFaithful-Refiner
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Faithful-RefinerLLM=Deepseek-V32025.05 | 17 | 90 | 2.53 | 20.18 | |
| Faithful-RefinerLLM=GPT-4o-mini2025.05 | 12 | 71 | 3.35 | 27.1 | |
| Faithful-RefinerLLM=Llama3.1-70b2025.05 | 11 | 68 | 2.9 | 25.4 | |
| Faithful-RefinerLLM=GPT-4o2025.05 | 10 | 79 | 3.22 | 22.32 | |
| Explanation-RefinerLLM=Llama3.1-70b2025.05 | 4 | 18 | 4.07 | 37.49 | |
| Explanation-RefinerLLM=GPT-4o2025.05 | 4 | 26 | 4.35 | 38.45 | |
| Explanation-RefinerLLM=Deepseek-V32025.05 | 4 | 38 | 3.71 | 35.97 | |
| Explanation-RefinerLLM=GPT-4o-mini2025.05 | 3 | 20 | 5.12 | 44.84 |