List your company

Question Answering on NQ Open (EM, Retrieval, Latency)

39.4Exact Match (EM)Self-Eval Gate
35.65636.62837.638.572Nov 12, 2025

Evaluation results

MethodLinks
Self-Eval GateBackbone=Qwen2.5-7B2025.1139.422.81.79
Self-RAG-liteBackbone=Qwen2.5-7B2025.113919.82.19
SKRBackbone=Qwen2.5-7B2025.11394.82.46
TARG-MarginBackbone=Qwen2.5-7B2025.11396.20.71
Never-RAGBackbone=Qwen2.5-7B2025.1138.803.29
FLAREBackbone=Qwen2.5-7B, θ=0.52025.1138.699.61.1
CRAG-liteBackbone=Qwen2.5-7B2025.1138202.01
Always-RAGBackbone=Qwen2.5-7B2025.1137.41002.92
FLAREBackbone=Qwen2.5-7B, θ=0.32025.1135.89614.4