Question Answering on PopQA (EM, Retrieval Rate, Latency)
23EMSKR
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SKRBackbone=Qwen2.5-7B2025.11 | 23 | 6.2 | 1.94 | |
| TARG-MarginBackbone=Qwen2.5-7B2025.11 | 22.6 | 0.1 | 0.45 | |
| Self-Eval GateBackbone=Qwen2.5-7B2025.11 | 22.2 | 0.2 | 1.16 | |
| CRAG-liteBackbone=Qwen2.5-7B2025.11 | 21.8 | 21 | 1.46 | |
| Self-RAG-liteBackbone=Qwen2.5-7B2025.11 | 21.8 | 8.8 | 1.87 | |
| Never-RAGBackbone=Qwen2.5-7B2025.11 | 20 | 0 | 2.13 | |
| Always-RAGBackbone=Qwen2.5-7B2025.11 | 14.6 | 100 | 3.83 | |
| FLAREBackbone=Qwen2.5-7B, θ=0.52025.11 | 12.8 | 100 | 1.73 | |
| FLAREBackbone=Qwen2.5-7B, θ=0.32025.11 | 10.6 | 99.2 | 0.77 |