Retrieval-Augmented Question Answering on Composite-9
94.4CorrectnessCustom ReAct
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Custom ReActBase Model=Qwen-2.5-7B2026.06 | 94.4 | 78 | 4.48 | |
| MonolithicBase Model=Qwen-2.5-7B2026.06 | 90 | 76 | 1.94 | |
| Regex-ScopedBase Model=Qwen-2.5-7B2026.06 | 90 | 80 | 2.64 | |
| HYBRID-ROUTEDBase Model=Qwen-2.5-7B2026.06 | 85.7 | 77 | 2.86 | |
| ColBERTv2 scoped + QwenBase Model=Qwen-2.5-7B2026.06 | 84 | 86 | 2.62 | |
| ColBERTv2 + QwenBase Model=Qwen-2.5-7B2026.06 | 82 | 78 | 1.8 | |
| MASDR-RAGBase Model=Qwen-2.5-7B2026.06 | 74 | 74 | 3.12 | |
| BM25 + QwenBase Model=Qwen-2.5-7B2026.06 | 74 | 58 | 1.74 | |
| SCOUT-RAGBase Model=Qwen-2.5-7B2026.06 | 66.7 | — | 23.6 | |
| MA-RAGBase Model=Qwen-2.5-7B2026.06 | 44.4 | — | 17.7 |