Multi-hop Question Answering on HotpotWikiQA 16k
25.4Overall ScoreLlama-3-8B-Instruct with DCS
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Llama-3-8B-Instruct with DCSBackbone=Llama-3-8B-Instruct, Context Strategy=DCS2025.06 | 25.4 | 29.07 | |
| Llama-3-8B-InstructBackbone=Llama-3-8B-Instruct, Context Strategy=None2025.06 | 14.22 | 24.22 | |
| Llama-3-8B-Instruct with StreamingBackbone=Llama-3-8B-Instruct, Context Strategy=Streaming2025.06 | 12.45 | 22.51 | |
| Llama-3-8B-Instruct with LM-infiniteBackbone=Llama-3-8B-Instruct, Context Strategy=LM-infinite2025.06 | 11.98 | 22.79 | |
| Llama-3-8B-Instruct with InfillmBackbone=Llama-3-8B-Instruct, Context Strategy=Infillm2025.06 | 7.74 | 23.58 |