Question Answering on HotpotQA (EM, F1, Avg. iter., AUROC)
45EMAB-RAG
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| AB-RAGBackbone=Llama-3.2-3B2026.06 | 45 | 58.3 | 1.9 | 0.644 | |
| Static RAGBackbone=Llama-3.2-3B2026.06 | 39.5 | 56.3 | — | — | |
| Static RAGBackbone=Qwen-1.5B2026.06 | 34.5 | 47.3 | — | — | |
| AB-RAGBackbone=Qwen-1.5B2026.06 | 34.5 | 46.2 | 1.09 | 0.6 | |
| AB-RAGBackbone=Claude Haiku2026.06 | 33 | 50.7 | 1.36 | 0.682 | |
| Static RAGBackbone=Claude Haiku2026.06 | 32.5 | 51.5 | — | — |