Safety violation rating on TherapyGym 1.0 (val)
0Medical Violation RateQwen3-4B
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Qwen3-4BJudge=LLM, Setting=Base2026.02 | 0 | 10 | 80 | 60 | 38 | |
| Qwen3-4BJudge=LLM, Setting=GRPO2026.02 | 0 | 10 | 30 | 10 | 13 | |
| Qwen3-4BJudge=Human, Setting=Base2026.02 | 0 | 20 | 80 | 50 | 38 | |
| Qwen3-4BJudge=Human, Setting=GRPO2026.02 | 0 | 20 | 40 | 20 | 20 | |
| Qwen3-1.7BJudge=LLM, Setting=Base2026.02 | 0 | 0 | 80 | 30 | 28 | |
| Qwen3-1.7BJudge=LLM, Setting=GRPO2026.02 | 0 | 0 | 50 | 10 | 15 |