Multi-hop Question Answering on Bamboogle offline Wiki-18 (test val)
53.4Exact Match (EM)ESearch-base
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ESearch-baseModel Scale=7B2025.10 | 53.4 | 65.6 | |
| StepSearch-baseModel Scale=7B2025.10 | 46.7 | 57.3 | |
| ESearch-instructModel Scale=7B2025.10 | 45.8 | 61.2 | |
| ESearch-instructModel Scale=3B2025.10 | 44.6 | 58.7 | |
| ZeroSearch-instructModel Scale=7B2025.10 | 43.3 | 54 | |
| Search-R1-baseModel Scale=7B2025.10 | 43 | 54.5 | |
| Research-instructModel Scale=7B2025.10 | 42.4 | 54.4 | |
| ZeroSearch-baseModel Scale=7B2025.10 | 41.7 | 53.2 | |
| ESearch-baseModel Scale=3B2025.10 | 41.4 | 52.9 | |
| R-Search-instruct-GRPOModel Scale=7B2025.10 | 40 | 51.7 | |
| StepSearch-instructModel Scale=7B2025.10 | 40 | 53.4 | |
| Search-R1-instructModel Scale=7B2025.10 | 38.4 | 50.1 | |
| R-Search-instruct-PPOModel Scale=7B2025.10 | 38.4 | 49.1 | |
| Research-baseModel Scale=7B2025.10 | 37.3 | 44.9 | |
| StepSearch-instructModel Scale=3B2025.10 | 34.4 | 45.2 | |
| SSRL-instructModel Scale=7B2025.10 | 34.4 | 46.6 | |
| StepSearch-baseModel Scale=3B2025.10 | 32.8 | 41.9 | |
| R-Search-instruct-PPOModel Scale=3B2025.10 | 26 | 35.5 | |
| Search-R1-instructModel Scale=3B2025.10 | 24 | 34.4 | |
| R-Search-instruct-GRPOModel Scale=3B2025.10 | 22.8 | 32.7 | |
| SSRL-instructModel Scale=3B2025.10 | 21.6 | 28.7 | |
| ZeroSearch-instructModel Scale=3B2025.10 | 19.3 | 29.9 | |
| Search-R1-baseModel Scale=3B2025.10 | 17.6 | 27 | |
| ZeroSearch-baseModel Scale=3B2025.10 | 16.5 | 25.6 |