List your company

Multi-hop Question Answering on Bamboogle (EM, RAG Improvement)

45.5Exact Match (EM)PyRAG
10.6619.70528.7537.795May 13, 2026

Evaluation results

MethodLinks
PyRAGBackbone=Qwen2.5-72B-Instruct2026.0545.523.9
PyRAGBackbone=Qwen2.5-7B-Instruct2026.0541.525.5
ITER-RETGENBackbone=Qwen2.5-72B-Instruct2026.0533.6—
ITER-RETGENBackbone=Qwen2.5-7B-Instruct2026.0529.6—
IRCoTBackbone=Qwen2.5-72B-Instruct2026.0528.3—
Self-AskBackbone=Qwen2.5-72B-Instruct2026.0526.1—
IRCoTBackbone=Qwen2.5-7B-Instruct2026.0524.3—
CoTBackbone=Qwen2.5-7B-Instruct2026.0523.2—
Self-AskBackbone=Qwen2.5-7B-Instruct2026.0522.1—
Vanilla RAGBackbone=Qwen2.5-72B-Instruct2026.0521.6—
CoTBackbone=Qwen2.5-72B-Instruct2026.0519.6—
Direct InferenceBackbone=Qwen2.5-72B-Instruct2026.0517.6—
Vanilla RAGBackbone=Qwen2.5-7B-Instruct2026.0516—
Direct InferenceBackbone=Qwen2.5-7B-Instruct2026.0512—