Multi-turn dialogue routing on ShareGPT-LF Qwen Series Set cross-domain (legal and financial)
91.46Success Rate (SR)MCTS Router
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MCTS RouterCandidate Set=Qwen Series Set2026.04 | 91.46 | 4.5 | |
| Greedy RouterCandidate Set=Qwen Series Set2026.04 | 87.18 | 4.53 | |
| DialRouterCandidate Set=Qwen Series Set2026.04 | 83.69 | 4.64 | |
| Qwen3-14BCandidate Set=Qwen Series Set2026.04 | 81.74 | 4.65 | |
| KNN RouterCandidate Set=Qwen Series Set2026.04 | 81.64 | 4.56 | |
| AvengersCandidate Set=Qwen Series Set2026.04 | 81.3 | 4.67 | |
| RandomCandidate Set=Qwen Series Set2026.04 | 81.03 | 4.67 | |
| Qwen3-8BCandidate Set=Qwen Series Set2026.04 | 80.66 | 4.68 | |
| Qwen3-4BCandidate Set=Qwen Series Set2026.04 | 80.24 | 4.8 |