Short-answer QA on NQ
62AccuracyDecoding_r
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Decoding_rLLM=Mistral-I7B, Defense=Decoding_r2024.05 | 62 | 29 | — | — | |
| VanillaLLM=Mistral-I7B, Defense=Vanilla2024.05 | 61 | — | — | — | |
| KeywordLLM=Mistral-I7B, Defense=Keyword2024.05 | 61 | 26 | — | — | |
| VanillaLLM=GPT-3.5, Retrieved passages (k)=10, Corruption size (k')=12024.05 | 58.8 | 0 | — | — | |
| VanillaLLM=Llama2-C7B, Defense=Vanilla2024.05 | 57 | — | — | — | |
| KeywordLLM=Llama2-C7B, Defense=Keyword2024.05 | 56 | 31 | — | — | |
| RobustRAG (Keyword)LLM=GPT-3.5, Retrieved passages (k)=10, Corruption size (k')=12024.05 | 54.2 | 37 | — | — | |
| Decoding_rLLM=Llama2-C7B, Defense=Decoding_r2024.05 | 53 | 36 | — | — | |
| No RAGLLM=Mistral-I7B, Defense=No RAG2024.05 | 30 | — | — | — | |
| No RAGLLM=GPT-3.5, Retrieved passages (k)=0, Corruption size (k')=12024.05 | 24.6 | — | — | — | |
| No RAGLLM=Llama2-C7B, Defense=No RAG2024.05 | 10 | — | — | — | |
| KeywordLLM=GPT-3.5, Attack=PIA, k=10, k'=12024.05 | — | — | 53 | 5.2 | |
| KeywordLLM=GPT-3.5, Attack=Poison, k=10, k'=12024.05 | — | — | 52.6 | 4.6 | |
| VanillaLLM=GPT-3.5, Attack=PIA, k=10, k'=12024.05 | — | — | 11 | 67.8 | |
| VanillaLLM=GPT-3.5, Attack=Poison, k=10, k'=12024.05 | — | — | 51.8 | 14.4 |