Multi-Hop Question Answering on Musique (F1, Collapse Point)
33.36F1 ScoreCalibAdv
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CalibAdvBackbone=Qwen2.5-7B-Base2026.04 | 33.36 | — | |
| LLDBackbone=Qwen2.5-7B-Base2026.04 | 31.57 | — | |
| CalibAdvBackbone=Qwen2.5-3B-Base2026.04 | 30.07 | — | |
| StepSearchBackbone=Qwen2.5-7B-Base2026.04 | 29.01 | — | |
| GiGPOBackbone=Qwen2.5-7B-Base2026.04 | 28.5 | — | |
| Search-R1Backbone=Qwen2.5-3B-Base2026.04 | 28.33 | 292 | |
| CalibAdvBackbone=Llama3.2-3B-Instruct2026.04 | 26.77 | — | |
| SimpleTIRBackbone=Qwen2.5-7B-Base2026.04 | 26.25 | 65 | |
| SimpleTIRBackbone=Qwen2.5-3B-Base2026.04 | 25.83 | 310 | |
| Search-R1Backbone=Qwen2.5-7B-Base2026.04 | 25.65 | 58 | |
| MT-GRPOBackbone=Qwen2.5-7B-Base2026.04 | 23.79 | — | |
| LLDBackbone=Llama3.2-3B-Instruct2026.04 | 23.7 | — | |
| SimpleTIRBackbone=Llama3.2-3B-Instruct2026.04 | 21.85 | 54 | |
| Search-R1Backbone=Llama3.2-3B-Instruct2026.04 | 19.92 | 40 |