Domain-specific Question Answering on QASC (Accuracy, Relative Accuracy Improvement (%))
68.36AccuracyVanilla Fine-tuning
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Vanilla Fine-tuningModel=Llama3-8B, Unlearning Target=none2025.05 | 68.36 | — | |
| Llama3.2-1BForgetting Task=none2025.05 | 42.98 | — | |
| LWFForgetting Task=mixed2025.05 | 5.54 | — | |
| LWFForgetting Task=dental2025.05 | 5.28 | — | |
| LWFForgetting Task=gsm8k2025.05 | 4.03 | — | |
| LWFForgetting Task=sst52025.05 | 3.02 | — | |
| LWFForgetting Task=psychol2025.05 | 2 | — | |
| LWFModel=Llama3-8B, Unlearning Target=gsm8k2025.05 | — | 5.37 | |
| LWFModel=Llama3-8B, Unlearning Target=sst52025.05 | — | 2.68 | |
| LWFModel=Llama3-8B, Unlearning Target=dental2025.05 | — | 1.26 | |
| LWFModel=Llama3-8B, Unlearning Target=psychol2025.05 | — | 4.42 | |
| LWFModel=Llama3-8B, Unlearning Target=mixed2025.05 | — | 7.9 |