Long Document Retrieval on NaturalQuestions (test)
59.98F1 ScoreAttentionRetriever
Evaluation Results
| Method | Links | |
|---|---|---|
| AttentionRetrieverBackbone=LLaMA-3.2 3B2026.02 | 59.98 | |
| AttentionRetrieverBackbone=Qwen-2.5 3B2026.02 | 56.55 | |
| GritLMModel Type=Dense2026.02 | 45.92 | |
| SPScannerModel Type=Autoregressive2026.02 | 42.37 | |
| GTE-Qwen2Model Type=Dense2026.02 | 41.31 | |
| GTRModel Type=Dense2026.02 | 41.1 | |
| Qwen3Model Type=Dense2026.02 | 40.91 | |
| ANCEModel Type=Dense2026.02 | 38.18 | |
| BM25Model Type=Sparse2026.02 | 30.55 | |
| DPRModel Type=Dense2026.02 | 29.81 | |
| CDEModel Type=Dense2026.02 | 24.16 |