Natural Language Explanation Generation on ECQA few-shot 60-shot
24.53AccuracySPARSEFIT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SPARSEFITFine-tuning Strategy=Att.Q+LN, Backbone=Llama 2-7B, Percentage Parameters=7.97%, Few-shot setting=60-shot2023.05 | 24.53 | 20.6 | |
| AdaLoRAFine-tuning Strategy=PEFT, Backbone=Llama 2-7B, Percentage Parameters=0.30%, Few-shot setting=60-shot2023.05 | 13.04 | 11.13 | |
| Full Fine-tuningFine-tuning Strategy=Baseline, Backbone=Llama 2-7B, Percentage Parameters=100%, Few-shot setting=60-shot2023.05 | 11.14 | 9.46 |