Multi-hop QA on NQ (test)
50.1EMParetoPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ParetoPOTool Use=Yes, Base Model=Qwen2.5-7B2026.06 | 50.1 | 0.9 | |
| ParetoPOTool Use=Yes, Base Model=Qwen2.5-3B2026.06 | 48 | 0.9 | |
| OTC-GRPOTool Use=Yes, Base Model=Qwen2.5-3B2026.06 | 44.4 | 1 | |
| OTC-GRPOTool Use=Yes, Base Model=Qwen2.5-7B2026.06 | 44.4 | 1 | |
| MO-GRPOTool Use=Yes, Base Model=Qwen2.5-7B2026.06 | 41.5 | 1.4 | |
| MO-GRPOTool Use=Yes, Base Model=Qwen2.5-3B2026.06 | 41.2 | 1.8 | |
| Search-R1Tool Use=Yes, Base Model=Qwen2.5-3B2026.06 | 40.4 | 1.4 | |
| Search-R1Tool Use=Yes, Base Model=Qwen2.5-7B2026.06 | 39.9 | 1.7 | |
| RAGTool Use=Yes, Base Model=Qwen2.5-7B2026.06 | 34.9 | 1 | |
| RAGTool Use=Yes, Base Model=Qwen2.5-3B2026.06 | 34.8 | 1 | |
| SFTTool Use=No, Base Model=Qwen2.5-7B2026.06 | 31.8 | — | |
| R1-BaseTool Use=No, Base Model=Qwen2.5-7B2026.06 | 27 | — | |
| SFTTool Use=No, Base Model=Qwen2.5-3B2026.06 | 24.9 | — | |
| R1-BaseTool Use=No, Base Model=Qwen2.5-3B2026.06 | 22.6 | — |