List your company

Multi-hop QA on 2WikiMultihopQA

56.4Exact Match (EM)SDGA-Phase
-0.59214.2042943.796Feb 28, 2026Mar 12, 2026Mar 24, 2026Apr 5, 2026Apr 17, 2026Apr 29, 2026May 12, 2026

Evaluation results

MethodLinks
SDGA-PhaseFramework=ZeroSearch, Model=Qwen2.5-7B, Search Depth (Smax)=52026.0556.4———
SDGA-AutoFramework=ZeroSearch, Model=Qwen2.5-7B, Search Depth (Smax)=52026.0555.8———
HippoRAGBackbone=Qwen2.5-72B-Instruct2026.0355.563.6——
GRPOmemory=none2026.0253.862.350.930.32
GraphRAG-RouterSystem=Our Method2026.0352.3———
TaSR-RAGBackbone=Qwen2.5-72B-Instruct2026.0351.866.2——
SDGA-AutoFramework=ZeroSearch, Model=Qwen2.5-3B, Search Depth (Smax)=52026.0551.8———
MemPO2026.0250.259.170.370.15
SDGA-PhaseFramework=ZeroSearch, Model=Qwen2.5-3B, Search Depth (Smax)=52026.0549.5———
GRPO-FullFramework=ZeroSearch, Model=Qwen2.5-7B2026.0547.8———
TaSR-RAGBackbone=Qwen2.5-7B-Instruct2026.0345.965.6——
Router-R1System=RL-based Agent2026.0345.6———
DeepResearcher2026.0243.951.441.240.34
LinearRAGSystem=RAG-Based2026.0343.5———
SDGA-PhaseFramework=Search-R1, Model=Qwen2.5-7B, Search Depth (Smax)=52026.0543.2———
GraphSearchSystem=Training-free Agent2026.0342.7———
GRPO-HalfFramework=ZeroSearch, Model=Qwen2.5-7B2026.0542.6———
HyperGraphRAGBackbone=Qwen2.5-72B-Instruct2026.0342.151.5——
ReSearch2026.0241.950.070.550.28
SDGA-AutoFramework=Search-R1, Model=Qwen2.5-7B, Search Depth (Smax)=52026.0541.6———
GRPO-FullFramework=ZeroSearch, Model=Qwen2.5-3B2026.0541.2———
Graph-R1System=RL-based Agent2026.0340.7———
SDGA-AutoFramework=Search-R1, Model=LLaMA-3.2-3B-Base, Search Depth (Smax)=52026.0538———
SDGA-PhaseFramework=Search-R1, Model=LLaMA-3.2-3B-Base, Search Depth (Smax)=52026.0537.6———
HippoRAG2System=RAG-Based2026.0337.4———
SDGA-AutoFramework=Search-R1, Model=Qwen2.5-3B, Search Depth (Smax)=52026.0537.3———
SDGA-AutoFramework=ZeroSearch, Model=LLaMA-3.2-3B-Base, Search Depth (Smax)=52026.0537.1———
SDGA-PhaseFramework=Search-R1, Model=Qwen2.5-3B, Search Depth (Smax)=52026.0537.1———
SDGA-PhaseFramework=ZeroSearch, Model=LLaMA-3.2-3B-Base, Search Depth (Smax)=52026.0536.8———
GRPO-FullFramework=Search-R1, Model=Qwen2.5-7B2026.0536.8———
GRPO-HalfFramework=ZeroSearch, Model=Qwen2.5-3B2026.0536.6———
qwen2.5mode=ReAct2026.0235.945.211.120.29
Vanilla RAGSystem=RAG-Based2026.0334.8———
GRPO-FullFramework=Search-R1, Model=Qwen2.5-3B2026.0533.7———
HyperGraphRAGSystem=RAG-Based2026.0333.6———
GraphRAGSystem=RAG-Based2026.0332.7———
GRPO-FullFramework=ZeroSearch, Model=LLaMA-3.2-3B-Base2026.0532.6———
GRPO-HalfFramework=Search-R1, Model=Qwen2.5-3B2026.0532.5———
GRPO-HalfFramework=Search-R1, Model=Qwen2.5-7B2026.0532.4———
RAPTORSystem=RAG-Based2026.0331.8———
GraphRAGBackbone=Qwen2.5-72B-Instruct2026.0330.938.5——
GRPO-FullFramework=Search-R1, Model=LLaMA-3.2-3B-Base2026.0530.6———
GraphRAGBackbone=Qwen2.5-7B-Instruct2026.0330.537.8——
Direct InferSystem=Basic LLM2026.0329.5———
CoTSystem=Basic LLM2026.0329.3———
GRPO-HalfFramework=ZeroSearch, Model=LLaMA-3.2-3B-Base2026.0528.7———
GRPO-HalfFramework=Search-R1, Model=LLaMA-3.2-3B-Base2026.0527.6———
Search-R1System=RL-based Agent2026.0326.9———
SDGA-AntiFramework=ZeroSearch, Model=Qwen2.5-7B, Search Depth (Smax)=52026.0526.8———
IRCoTBackbone=Qwen2.5-72B-Instruct2026.0325.132.1——
SDGA-AntiFramework=ZeroSearch, Model=Qwen2.5-3B, Search Depth (Smax)=52026.0524.7———
StructRAGBackbone=Qwen2.5-72B-Instruct2026.0324.537.2——
SDGA-AntiFramework=Search-R1, Model=Qwen2.5-7B, Search Depth (Smax)=52026.0524.2———
Search-o1System=Training-free Agent2026.0323———
RAGBackbone=Qwen2.5-72B-Instruct2026.0322.729.1——
CoTBackbone=Qwen2.5-72B-Instruct2026.0320.526.2——
HippoRAGBackbone=Qwen2.5-7B-Instruct2026.0319.738.2——
SDGA-AntiFramework=Search-R1, Model=Qwen2.5-3B, Search Depth (Smax)=52026.0518.3———
StructRAGBackbone=Qwen2.5-7B-Instruct2026.0316.723.9——
SDGA-AntiFramework=ZeroSearch, Model=LLaMA-3.2-3B-Base, Search Depth (Smax)=52026.0516.4———
SDGA-AntiFramework=Search-R1, Model=LLaMA-3.2-3B-Base, Search Depth (Smax)=52026.0515.7———
Direct InferenceBackbone=Qwen2.5-72B-Instruct2026.0315.319.6——
IRCoTBackbone=Qwen2.5-7B-Instruct2026.0314.921.3——
RAGBackbone=Qwen2.5-7B-Instruct2026.0313.547.7——
Direct InferenceBackbone=Qwen2.5-7B-Instruct2026.0312.630.1——
CoTBackbone=Qwen2.5-7B-Instruct2026.0310.815.1——
HypergraphRAGBackbone=Qwen2.5-7B-Instruct2026.031.61.9——