Multi-hop Question Answering on HotpotQA 10 related documents
60.61F1Oracle
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| OracleBackbone=LongChat-13B2024.06 | 60.61 | 42.2 | — | |
| OracleBackbone=LLAMA-2-7B2024.06 | 57.68 | 42.2 | — | |
| QGCBackbone=LongChat-13B2024.06 | 52.12 | 8.8 | 1.849 | |
| QGCBackbone=LLAMA-2-7B2024.06 | 51.64 | 8.8 | 0.927 | |
| QGCBackbone=LLAMA-2-7B, epsilon=0.422024.06 | 51.23 | 12.1 | — | |
| Cond.PPLBackbone=LongChat-13B2024.06 | 49.82 | 4.3 | 1.267 | |
| BGE-RerankerBackbone=LongChat-13B2024.06 | 47.96 | 4.4 | 1.689 | |
| Cond.PPLBackbone=LLAMA-2-7B2024.06 | 47.84 | 4.3 | 0.623 | |
| BGE-RerankerBackbone=LLAMA-2-7B2024.06 | 46.13 | 4.4 | 0.724 | |
| LongLLMLinguaBackbone=LongChat-13B2024.06 | 45.43 | 3.8 | 0.683 | |
| Original PromptBackbone=LongChat-13B2024.06 | 44.76 | 1 | — | |
| Original PromptBackbone=LLAMA-2-7B2024.06 | 44.24 | 1 | — | |
| Sentence-BERTBackbone=LongChat-13B2024.06 | 42.92 | 4.4 | 1.93 | |
| Sentence-BERTBackbone=LLAMA-2-7B2024.06 | 40.65 | 4.4 | 0.752 | |
| LongLLMLinguaBackbone=LLAMA-2-7B2024.06 | 39.87 | 3.8 | 0.438 | |
| ICAEBackbone=LLAMA-2-7B2024.06 | 34.5 | 9.5 | — | |
| Selective-ContextBackbone=LongChat-13B2024.06 | 29.68 | 2.6 | 0.456 | |
| AutoCompressorBackbone=LLAMA-2-7B2024.06 | 29.02 | 8.1 | 0.833 | |
| Selective-ContextBackbone=LLAMA-2-7B2024.06 | 28.21 | 2.6 | 0.332 | |
| Closed-bookBackbone=LongChat-13B2024.06 | 22.19 | — | — | |
| Closed-bookBackbone=LLAMA-2-7B2024.06 | 10.54 | — | — |