Skill retrieval on LogicBench
31.4nDCG@1Qwen3-32B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen3-32BRetrieval Stage=Reranker, Candidate selection=top-50 candidates retrieved by BM252026.04 | 31.4 | 44.2 | |
| Qwen3-235BRetrieval Stage=Reranker, Candidate selection=top-50 candidates retrieved by BM252026.04 | 30.9 | 44.4 | |
| Llama-3.3-70BRetrieval Stage=Reranker, Candidate selection=top-50 candidates retrieved by BM252026.04 | 27.4 | 41.8 | |
| Mistral3.1-24BRetrieval Stage=Reranker, Candidate selection=top-50 candidates retrieved by BM252026.04 | 26.7 | 40.3 | |
| Qwen3-4BRetrieval Stage=Reranker, Candidate selection=top-50 candidates retrieved by BM252026.04 | 21.8 | 32.3 | |
| Llama-3.1-8BRetrieval Stage=Reranker, Candidate selection=top-50 candidates retrieved by BM252026.04 | 15.9 | 28.8 | |
| BM25Retrieval Stage=First-stage2026.04 | 12 | 22.6 | |
| HybridRetrieval Stage=First-stage2026.04 | 12 | 21 | |
| ContrieverRetrieval Stage=First-stage2026.04 | 5.5 | 10.9 | |
| BGERetrieval Stage=First-stage2026.04 | 4.1 | 11.1 | |
| TF-IDFRetrieval Stage=First-stage2026.04 | 1.8 | 8.3 |