Multi-hop Question Answering on HotpotQA offline Wiki-18 (test val)
44.7EMESearch-instruct
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ESearch-instructModel Scale=3B2025.10 | 44.7 | 58.7 | |
| ESearch-instructModel Scale=7B2025.10 | 44.2 | 57.6 | |
| ESearch-baseModel Scale=7B2025.10 | 43.4 | 56.4 | |
| Search-R1-baseModel Scale=7B2025.10 | 43.2 | 54.7 | |
| ESearch-baseModel Scale=3B2025.10 | 41.5 | 54.8 | |
| Search-R1-instructModel Scale=7B2025.10 | 39.4 | 50.2 | |
| R-Search-instruct-GRPOModel Scale=7B2025.10 | 39.1 | 50 | |
| ZeroSearch-instructModel Scale=7B2025.10 | 38.8 | 49.7 | |
| StepSearch-instructModel Scale=7B2025.10 | 38.6 | 50.2 | |
| SSRL-instructModel Scale=7B2025.10 | 38 | 48.9 | |
| StepSearch-baseModel Scale=7B2025.10 | 38 | 49.3 | |
| ZeroSearch-baseModel Scale=7B2025.10 | 37.5 | 48.1 | |
| Research-instructModel Scale=7B2025.10 | 36.2 | 47.1 | |
| StepSearch-instructModel Scale=3B2025.10 | 34.5 | 45.2 | |
| R-Search-instruct-PPOModel Scale=7B2025.10 | 33.8 | 43.9 | |
| R-Search-instruct-GRPOModel Scale=3B2025.10 | 32.9 | 42.7 | |
| StepSearch-baseModel Scale=3B2025.10 | 32.9 | 43.4 | |
| SSRL-instructModel Scale=3B2025.10 | 31.4 | 40.8 | |
| Search-R1-instructModel Scale=3B2025.10 | 30.4 | 40.1 | |
| Research-baseModel Scale=7B2025.10 | 29.4 | 38.8 | |
| R-Search-instruct-PPOModel Scale=3B2025.10 | 28.9 | 38.1 | |
| ZeroSearch-baseModel Scale=3B2025.10 | 28.1 | 37.7 | |
| Search-R1-baseModel Scale=3B2025.10 | 27.2 | 36.1 | |
| ZeroSearch-instructModel Scale=3B2025.10 | 26.7 | 35.3 |