Multi-hop Question Answering on MuSiQue v1.0 (test)
2.3Task latency (s)No Retrieval
Evaluation Results
| Method | Links | |
|---|---|---|
| No RetrievalBackbone=Llama3.1-8B2025.03 | 2.3 | |
| VanillaRAGBackbone=Llama3.1-8B, k=42025.03 | 2.55 | |
| L-RAGBackbone=Llama3.1-8B, k=42025.03 | 2.62 | |
| VanillaRAGBackbone=Llama3.1-8B, k=82025.03 | 2.67 | |
| L-RAGBackbone=Llama3.1-8B, k=82025.03 | 2.86 | |
| HyDEBackbone=Llama3.1-8B2025.03 | 4.81 | |
| IRCoTBackbone=Llama3.1-8B2025.03 | 14.57 | |
| SelfAskBackbone=Llama3.1-8B2025.03 | 19.62 | |
| IterRetGenBackbone=Llama3.1-8B2025.03 | 21.18 |