Multi-hop Question Answering on HotpotQA first 1000 samples (dev)
39.7EMPathfinder
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| PathfinderBase Model=Qwen-2.5-7B-Instruct, Distribution Setting=out-of-distribution2025.12 | 39.7 | 41.9 | 51.88 | |
| DeepRAG (Imi-retrieve only)Base Model=Qwen-2.5-7B-Instruct, Distribution Setting=out-of-distribution2025.12 | 39.1 | 41.7 | 50.81 | |
| PathfinderBase Model=Gemma-2-9B-Instruct, Distribution Setting=out-of-distribution2025.12 | 36.6 | 39.4 | 47.91 | |
| DeepRAG (Imi-retrieve only)Base Model=Gemma-2-9B-Instruct, Distribution Setting=out-of-distribution2025.12 | 36.4 | 39 | 47.56 | |
| DeepRAG (Imi)Base Model=Gemma-2-9B-Instruct, Distribution Setting=out-of-distribution2025.12 | 32.8 | 34.9 | 44 | |
| DeepRAG (Imi)Base Model=Qwen-2.5-7B-Instruct, Distribution Setting=out-of-distribution2025.12 | 27.8 | 29.5 | 37.72 |