Multi-Turn User Simulation on PEARL (test)
95.31Success Rate (SR)UserSim-Qwen 8B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| UserSim-Qwen 8BMode=Fine-tuned, Parameters=8B2026.03 | 95.31 | 0.0232 | 2.37 | |
| UserSim-Llama 8BMode=Fine-tuned, Parameters=8B2026.03 | 92.65 | 0.0498 | 2.37 | |
| Qwen3 32BMode=Zero-shot, Parameters=32B2026.03 | 77.72 | 0.0097 | 21.31 | |
| Llama3.1 70BMode=Zero-shot, Parameters=70B2026.03 | 36.15 | 0.0033 | 63.51 |