Multi-hop Question Answering on HotpotQA (F1, Latency)
86.69F1 ScoreFlowCompile
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FlowCompileSelection Priority=Accuracy-priority2026.05 | 86.69 | 7.8 | |
| Qwen3-8BBackbone=Qwen3-8B2026.05 | 86.58 | 18.5 | |
| Full BaselineBackbone=Qwen3-14B2026.05 | 86.29 | 26.5 | |
| FlowCompileSelection Priority=Latency-priority2026.05 | 85.32 | 3.8 | |
| Qwen3-4BBackbone=Qwen3-4B2026.05 | 83.89 | 12.5 | |
| MaASBackbone=Qwen3-8B2026.05 | 83.44 | 14.1 | |
| KNN Router2026.05 | 73.75 | 6.9 | |
| MaASBackbone=Qwen3-4B2026.05 | 71.82 | 6.3 |