Open-domain Question Answering on PopQA
47.4AccuracySAAS
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SAASBackbone=Qwen2.5-7B-Instruct2026.05 | 47.4 | 1.01 | |
| RFTBackbone=Qwen2.5-7B-Instruct2026.05 | 47.3 | 1.46 | |
| HiPRAGBackbone=Qwen2.5-7B-Instruct2026.05 | 46.8 | 2.03 | |
| SAASBackbone=Qwen2.5-3B-Instruct2026.05 | 45.1 | 1.01 | |
| Search-R1Backbone=Qwen2.5-7B-Instruct2026.05 | 44.9 | 1.26 | |
| HiPRAGBackbone=Qwen2.5-3B-Instruct2026.05 | 43.8 | 1.63 | |
| StepSearchBackbone=Qwen2.5-7B-Instruct2026.05 | 43.6 | 1.17 | |
| Search-R1Backbone=Qwen2.5-3B-Instruct2026.05 | 42.4 | 1.23 | |
| RFTBackbone=Qwen2.5-3B-Instruct2026.05 | 42.3 | 1.81 | |
| StepSearchBackbone=Qwen2.5-3B-Instruct2026.05 | 39.6 | 1.37 | |
| Direct InferenceBackbone=Qwen2.5-7B-Instruct2026.05 | 16.9 | — | |
| Direct InferenceBackbone=Qwen2.5-3B-Instruct2026.05 | 14.4 | — |