Question Answering on TriviaQA adversarial Contriever retrieved top 10
57.72EMQGC
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| QGCBackbone=LongChat-13B2024.06 | 57.72 | 7.9 | 1.832 | |
| QGCBackbone=LLAMA-2-7B2024.06 | 57.46 | 7.9 | 0.902 | |
| QGCBackbone=LLAMA-2-7B, epsilon=0.422024.06 | 57.11 | 10.9 | — | |
| Cond.PPLBackbone=LLAMA-2-7B2024.06 | 52.76 | 4.5 | 0.602 | |
| Cond.PPLBackbone=LongChat-13B2024.06 | 52.48 | 4.5 | 1.287 | |
| LongLLMLinguaBackbone=LongChat-13B2024.06 | 51.51 | 3.7 | 0.724 | |
| LongLLMLinguaBackbone=LLAMA-2-7B2024.06 | 50.44 | 3.7 | 0.432 | |
| Sentence-BERTBackbone=LLAMA-2-7B2024.06 | 49.49 | 4.5 | 0.731 | |
| Original PromptBackbone=LLAMA-2-7B2024.06 | 49.47 | 1 | — | |
| ICAEBackbone=LLAMA-2-7B2024.06 | 48.91 | 10.2 | — | |
| Sentence-BERTBackbone=LongChat-13B2024.06 | 48.89 | 4.5 | 1.957 | |
| Original PromptBackbone=LongChat-13B2024.06 | 48.7 | 1 | — | |
| BGE-RerankerBackbone=LLAMA-2-7B2024.06 | 48.69 | 4.5 | 0.683 | |
| Selective-ContextBackbone=LLAMA-2-7B2024.06 | 48.55 | 2.5 | 0.303 | |
| BGE-RerankerBackbone=LongChat-13B2024.06 | 47.71 | 4.5 | 1.724 | |
| Selective-ContextBackbone=LongChat-13B2024.06 | 42.73 | 2.5 | 0.465 | |
| Closed-bookBackbone=LongChat-13B2024.06 | 36.07 | — | — | |
| Closed-bookBackbone=LLAMA-2-7B2024.06 | 30.7 | — | — | |
| AutoCompressorBackbone=LLAMA-2-7B2024.06 | 29.17 | 8.7 | 0.823 |