Multi-hop Reasoning on 2WikiMQA IRCoT 500 samples (test)
52.8ACCActiShade
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ActiShadeBackbone=Qwen2.5-7B-Instruct, Retrieval Mode=Active/Iterative retrieval2026.01 | 52.8 | 50.47 | |
| ActiShadeBackbone=Qwen2.5-14B-Instruct, Retrieval Mode=Active/Iterative retrieval2026.01 | 51.13 | 53.29 | |
| DRAGINBackbone=Qwen2.5-14B-Instruct, Retrieval Mode=Multi-round retrieval2026.01 | 48.1 | 49.87 | |
| SelfASKBackbone=Qwen2.5-14B-Instruct, Retrieval Mode=Multi-round retrieval2026.01 | 47.49 | 49.13 | |
| DRAGINBackbone=Qwen2.5-7B-Instruct, Retrieval Mode=Multi-round retrieval2026.01 | 45.9 | 45.87 | |
| ActiShadeBackbone=Llama-3-8B-Instruct, Retrieval Mode=Active/Iterative retrieval2026.01 | 45.8 | 46.02 | |
| IRCoTBackbone=Qwen2.5-14B-Instruct, Retrieval Mode=Multi-round retrieval2026.01 | 44.6 | 45.71 | |
| Iter-RetGenBackbone=Qwen2.5-14B-Instruct, Retrieval Mode=Multi-round retrieval2026.01 | 43.2 | 45.19 | |
| SelfASKBackbone=Qwen2.5-7B-Instruct, Retrieval Mode=Multi-round retrieval2026.01 | 43.17 | 44.3 | |
| FLAREBackbone=Llama-3-8B-Instruct, Retrieval Mode=Multi-round retrieval2026.01 | 41.35 | 42.24 | |
| Iter-RetGenBackbone=Qwen2.5-7B-Instruct, Retrieval Mode=Multi-round retrieval2026.01 | 41.2 | 42.98 | |
| DRAGINBackbone=Llama-3-8B-Instruct, Retrieval Mode=Multi-round retrieval2026.01 | 40.78 | 42.31 | |
| IRCoTBackbone=Qwen2.5-7B-Instruct, Retrieval Mode=Multi-round retrieval2026.01 | 40.21 | 40.84 | |
| FLAREBackbone=Qwen2.5-14B-Instruct, Retrieval Mode=Multi-round retrieval2026.01 | 40.01 | 40.74 | |
| SelfASKBackbone=Llama-3-8B-Instruct, Retrieval Mode=Multi-round retrieval2026.01 | 39.5 | 43.87 | |
| FLAREBackbone=Qwen2.5-7B-Instruct, Retrieval Mode=Multi-round retrieval2026.01 | 39.21 | 40.8 | |
| Direct-RBackbone=Qwen2.5-14B-Instruct, Retrieval Mode=Single-round retrieval2026.01 | 39 | 43.14 | |
| Iter-RetGenBackbone=Llama-3-8B-Instruct, Retrieval Mode=Multi-round retrieval2026.01 | 38.71 | 44.56 | |
| Direct-RBackbone=Qwen2.5-7B-Instruct, Retrieval Mode=Single-round retrieval2026.01 | 38.6 | 41.17 | |
| IRCoTBackbone=Llama-3-8B-Instruct, Retrieval Mode=Multi-round retrieval2026.01 | 34.2 | 41.01 | |
| CoTBackbone=Qwen2.5-14B-Instruct, Retrieval Mode=None2026.01 | 33.4 | 35.76 | |
| CoTBackbone=Qwen2.5-7B-Instruct, Retrieval Mode=None2026.01 | 29 | 32.24 | |
| DirectBackbone=Qwen2.5-14B-Instruct, Retrieval Mode=None2026.01 | 29 | 32.39 | |
| Direct-RBackbone=Llama-3-8B-Instruct, Retrieval Mode=Single-round retrieval2026.01 | 28.37 | 35.56 | |
| CoTBackbone=Llama-3-8B-Instruct, Retrieval Mode=None2026.01 | 27.6 | 34.19 | |
| DirectBackbone=Qwen2.5-7B-Instruct, Retrieval Mode=None2026.01 | 26.8 | 29.95 | |
| DirectBackbone=Llama-3-8B-Instruct, Retrieval Mode=None2026.01 | 26.6 | 31.25 |