Question Answering on Knowns.QA (1,000 samples subset of COUNTERFACT)
14.32Misleading RateICL
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ICLstrategy=In-context learning2025.02 | 14.32 | 77.21 | |
| FT-Llama-Lorafine-tuning=LoRA2025.02 | 37.47 | 90.45 | |
| LLMmodel=Llama-7B-Chat2025.02 | 42.24 | 93.56 | |
| FT-Llama-Fullfine-tuning=Full2025.02 | 43.15 | 91.23 | |
| System Prompt2025.02 | 45.82 | 96.18 | |
| COTstrategy=Chain-of-thought2025.02 | 47.61 | 94.57 | |
| Astute-RAG2025.02 | 51.43 | 87.94 | |
| Grfttuning=Gated ReFT2025.02 | 62.52 | 95.11 | |
| Grft-requerytuning=Gated ReFT, mechanism=requery2025.02 | 75.78 | 97.61 |