Multi-Hop Question Answering on 2WikiMultihopQA (F1/Collapse Point)
56.91F1 ScoreCalibAdv
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CalibAdvBackbone=Qwen2.5-7B-Base2026.04 | 56.91 | — | |
| LLDBackbone=Qwen2.5-7B-Base2026.04 | 56.17 | — | |
| Search-R1Backbone=Qwen2.5-7B-Base2026.04 | 53 | 58 | |
| CalibAdvBackbone=Qwen2.5-3B-Base2026.04 | 52.51 | — | |
| SimpleTIRBackbone=Qwen2.5-7B-Base2026.04 | 49.95 | 65 | |
| Search-R1Backbone=Qwen2.5-3B-Base2026.04 | 49.83 | 292 | |
| SimpleTIRBackbone=Qwen2.5-3B-Base2026.04 | 48.75 | 310 | |
| LLDBackbone=Llama3.2-3B-Instruct2026.04 | 47.8 | — | |
| GiGPOBackbone=Qwen2.5-7B-Base2026.04 | 47.79 | — | |
| CalibAdvBackbone=Llama3.2-3B-Instruct2026.04 | 47.36 | — | |
| StepSearchBackbone=Qwen2.5-7B-Base2026.04 | 46.94 | — | |
| MT-GRPOBackbone=Qwen2.5-7B-Base2026.04 | 46.46 | — | |
| SimpleTIRBackbone=Llama3.2-3B-Instruct2026.04 | 42.84 | 54 | |
| Search-R1Backbone=Llama3.2-3B-Instruct2026.04 | 40.19 | 40 |