Long document retrieval on QASA (test)
56.63F1 ScoreAttentionRetriever
Evaluation Results
| Method | Links | |
|---|---|---|
| AttentionRetrieverBackbone=Qwen-2.5 3B2026.02 | 56.63 | |
| AttentionRetrieverBackbone=LLaMA-3.2 3B2026.02 | 55.84 | |
| SPScannerModel Type=Autoregressive2026.02 | 46.04 | |
| Qwen3Model Type=Dense2026.02 | 44.14 | |
| GritLMModel Type=Dense2026.02 | 43.94 | |
| GTRModel Type=Dense2026.02 | 39.77 | |
| BM25Model Type=Sparse2026.02 | 37.95 | |
| GTE-Qwen2Model Type=Dense2026.02 | 37.85 | |
| ANCEModel Type=Dense2026.02 | 37.24 | |
| DPRModel Type=Dense2026.02 | 21.81 | |
| CDEModel Type=Dense2026.02 | 13.41 |