Multi-Hop QA on Bamboogle (Exact Match (EM))
57.8Exact Match (EM)SDGA-Phase
Evaluation Results
| Method | Links | |
|---|---|---|
| SDGA-PhaseFramework=ZeroSearch, Model=Qwen2.5-7B, Search Depth (Smax)=52026.05 | 57.8 | |
| SDGA-PhaseFramework=ZeroSearch, Model=Qwen2.5-3B, Search Depth (Smax)=52026.05 | 51.4 | |
| SDGA-AutoFramework=ZeroSearch, Model=Qwen2.5-7B, Search Depth (Smax)=52026.05 | 47.4 | |
| GRPO-FullFramework=ZeroSearch, Model=Qwen2.5-7B2026.05 | 46.8 | |
| GRPO-FullFramework=ZeroSearch, Model=Qwen2.5-3B2026.05 | 39.8 | |
| ReSeekBackbone=Qwen2.5-7b-Instruct2025.10 | 39.2 | |
| SDGA-AutoFramework=ZeroSearch, Model=Qwen2.5-3B, Search Depth (Smax)=52026.05 | 39.1 | |
| SDGA-PhaseFramework=Search-R1, Model=Qwen2.5-7B, Search Depth (Smax)=52026.05 | 38.4 | |
| Search-R1Backbone=Qwen2.5-7b-Instruct2025.10 | 36.8 | |
| SDGA-PhaseFramework=Search-R1, Model=Qwen2.5-3B, Search Depth (Smax)=52026.05 | 32.3 | |
| ReSeekBackbone=Qwen2.5-3b-Instruct2025.10 | 30.4 | |
| Search-o1Backbone=Qwen2.5-7b-Instruct2025.10 | 29.6 | |
| GRPO-HalfFramework=ZeroSearch, Model=Qwen2.5-7B2026.05 | 29.6 | |
| R1Backbone=Qwen2.5-7b-Instruct2025.10 | 29.3 | |
| ZeroSearchBackbone=Qwen2.5-7b-Instruct2025.10 | 27.8 | |
| Search-R1Backbone=Qwen2.5-3b-Instruct2025.10 | 26.4 | |
| SDGA-AutoFramework=Search-R1, Model=Qwen2.5-7B, Search Depth (Smax)=52026.05 | 24.6 | |
| GRPO-HalfFramework=ZeroSearch, Model=Qwen2.5-3B2026.05 | 24.1 | |
| GRPO-FullFramework=Search-R1, Model=Qwen2.5-7B2026.05 | 23.6 | |
| SDGA-AutoFramework=Search-R1, Model=LLaMA-3.2-3B-Base, Search Depth (Smax)=52026.05 | 23.4 | |
| CoTBackbone=Qwen2.5-7b-Instruct2025.10 | 23.2 | |
| SDGA-PhaseFramework=Search-R1, Model=LLaMA-3.2-3B-Base, Search Depth (Smax)=52026.05 | 22.9 | |
| SDGA-AutoFramework=ZeroSearch, Model=LLaMA-3.2-3B-Base, Search Depth (Smax)=52026.05 | 22.1 | |
| SDGA-PhaseFramework=ZeroSearch, Model=LLaMA-3.2-3B-Base, Search Depth (Smax)=52026.05 | 21.8 | |
| SDGA-AntiFramework=ZeroSearch, Model=Qwen2.5-7B, Search Depth (Smax)=52026.05 | 21.4 | |
| SDGA-AutoFramework=Search-R1, Model=Qwen2.5-3B, Search Depth (Smax)=52026.05 | 21.4 | |
| SDGA-AntiFramework=ZeroSearch, Model=Qwen2.5-3B, Search Depth (Smax)=52026.05 | 20.9 | |
| GRPO-HalfFramework=Search-R1, Model=Qwen2.5-3B2026.05 | 20.9 | |
| RAGBackbone=Qwen2.5-7b-Instruct2025.10 | 20.8 | |
| GRPO-FullFramework=Search-R1, Model=Qwen2.5-3B2026.05 | 20.1 | |
| R1Backbone=Qwen2.5-3b-Instruct2025.10 | 19.2 | |
| SDGA-AntiFramework=Search-R1, Model=Qwen2.5-7B, Search Depth (Smax)=52026.05 | 16.8 | |
| GRPO-FullFramework=ZeroSearch, Model=LLaMA-3.2-3B-Base2026.05 | 16.4 | |
| GRPO-FullFramework=Search-R1, Model=LLaMA-3.2-3B-Base2026.05 | 15.2 | |
| GRPO-HalfFramework=ZeroSearch, Model=LLaMA-3.2-3B-Base2026.05 | 13.6 | |
| GRPO-HalfFramework=Search-R1, Model=LLaMA-3.2-3B-Base2026.05 | 12.7 | |
| GRPO-HalfFramework=Search-R1, Model=Qwen2.5-7B2026.05 | 12.4 | |
| Direct InferenceBackbone=Qwen2.5-7b-Instruct2025.10 | 12 | |
| ZeroSearchBackbone=Qwen2.5-3b-Instruct2025.10 | 11.1 | |
| SDGA-AntiFramework=Search-R1, Model=Qwen2.5-3B, Search Depth (Smax)=52026.05 | 10.1 | |
| SDGA-AntiFramework=ZeroSearch, Model=LLaMA-3.2-3B-Base, Search Depth (Smax)=52026.05 | 8.5 | |
| SDGA-AntiFramework=Search-R1, Model=LLaMA-3.2-3B-Base, Search Depth (Smax)=52026.05 | 8.3 | |
| RAGBackbone=Qwen2.5-3b-Instruct2025.10 | 8 | |
| Direct InferenceBackbone=Qwen2.5-3b-Instruct2025.10 | 2.4 | |
| Search-o1Backbone=Qwen2.5-3b-Instruct2025.10 | 1.9 | |
| CoTBackbone=Qwen2.5-3b-Instruct2025.10 | 0 |