Logical Refinement of Natural Language Explanations on e-SNLI
41Initial PerformanceFaithful-Refiner
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Faithful-RefinerLLM=Deepseek-V32025.05 | 41 | 95 | 1.5 | 9.52 | |
| Faithful-RefinerLLM=GPT-4o2025.05 | 39 | 89 | 1.54 | 10.24 | |
| Faithful-RefinerLLM=Llama3.1-70b2025.05 | 36 | 78 | 2.38 | 16.28 | |
| Faithful-RefinerLLM=GPT-4o-mini2025.05 | 32 | 77 | 2.27 | 16.62 | |
| Explanation-RefinerLLM=GPT-4o2025.05 | 31 | 71 | 3.62 | 32.34 | |
| Explanation-RefinerLLM=Deepseek-V32025.05 | 25 | 69 | 2.82 | 27.74 | |
| Explanation-RefinerLLM=Llama3.1-70b2025.05 | 23 | 51 | 4.08 | 34.56 | |
| Explanation-RefinerLLM=GPT-4o-mini2025.05 | 13 | 30 | 3.65 | 32.55 |