Question Answering on ConFiQA-MR (Held-out)
57.3AccuracyRAG
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RAGBackbone=Llama-3.1-8B, Setting=Zero-shot2026.05 | 57.3 | 59.3 | — | — | |
| LODO SABERBackbone=Llama-3.1-8B, Setting=Zero-shot2026.05 | 56.9 | 74.3 | -0.4 | 15 | |
| LODO SABERBackbone=Qwen-2.5-3B, Setting=Zero-shot2026.05 | 55.8 | 76.6 | 2 | 19.2 | |
| LODO SABERBackbone=Qwen-2.5-7B, Setting=Zero-shot2026.05 | 55.2 | 81.9 | 15 | 16.7 | |
| Imp-SCRBackbone=Qwen-2.5-3B, Setting=Zero-shot2026.05 | 53.8 | 57.4 | — | — | |
| LODO SABERBackbone=Llama-3.2-3B, Setting=Zero-shot2026.05 | 51.6 | 78.7 | 6.8 | 20.2 | |
| Ctx-EvalBackbone=Llama-3.2-3B, Setting=Zero-shot2026.05 | 44.8 | 58.5 | — | — | |
| Ctx-EvalBackbone=Qwen-2.5-7B, Setting=Zero-shot2026.05 | 40.2 | 65.2 | — | — |