Downstream Generation and Ranking Alignment on MKQA (test)
49.93-gram RecallBGE-Reranker + LAURA
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BGE-Reranker + LAURAGenerator Model=Llama, Top-k candidates=52026.04 | 49.9 | 0.236 | |
| Qwen-Reranker + LAURAGenerator Model=Llama, Top-k candidates=52026.04 | 49.2 | 0.269 | |
| BGE-RerankerGenerator Model=Llama, Top-k candidates=52026.04 | 48.9 | 0.198 | |
| BGE-Reranker + LAURAGenerator Model=Qwen, Top-k candidates=52026.04 | 47.7 | 0.247 | |
| Qwen-RerankerGenerator Model=Llama, Top-k candidates=52026.04 | 47.1 | 0.129 | |
| BGE-RerankerGenerator Model=Qwen, Top-k candidates=52026.04 | 46.7 | 0.188 | |
| Qwen-Reranker + LAURAGenerator Model=Qwen, Top-k candidates=52026.04 | 46.7 | 0.264 | |
| Qwen-RerankerGenerator Model=Qwen, Top-k candidates=52026.04 | 44.9 | 0.127 |