Question Answering on NQ (EM, ACC)
34.3EMSkill-RAG
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Skill-RAGBackbone=Gemma2-9B2026.04 | 34.3 | 49.7 | |
| DRAGINBackbone=Gemma2-9B2026.04 | 34.2 | 38.9 | |
| Probing-RAGBackbone=Gemma2-9B2026.04 | 34.1 | 48.5 | |
| FLAREBackbone=Gemma2-9B2026.04 | 29.7 | 36.4 | |
| Adaptive-RAGBackbone=Gemma2-9B2026.04 | 29.4 | 34.5 | |
| No RetrievalBackbone=Gemma2-9B2026.04 | 27.3 | 42.7 | |
| Single-step ApproachBackbone=Gemma2-9B2026.04 | 26.1 | 40.7 |