CBT skill scoring on TherapyGym 1.0 (val)
52Agency ScoreQwen3-4B
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Qwen3-4BJudge=Human, Setting=GRPO2026.02 | 52 | 65 | 68 | 72 | 62 | 65 | 67 | 58 | 27 | 60 | |
| Qwen3-4BJudge=LLM, Setting=GRPO2026.02 | 42 | 65 | 77 | 78 | 68 | 61 | 70 | 57 | 17 | 59 | |
| Qwen3-1.7BJudge=LLM, Setting=GRPO2026.02 | 7 | 30 | 42 | 42 | 23 | 25 | 40 | 33 | 18 | 29 | |
| Qwen3-4BJudge=Human, Setting=Base2026.02 | 2 | 4 | 17 | 37 | 15 | 11 | 4 | 2 | 0 | 10 | |
| Qwen3-4BJudge=LLM, Setting=Base2026.02 | 0 | 3 | 43 | 58 | 15 | 10 | 10 | 3 | 0 | 16 | |
| Qwen3-1.7BJudge=LLM, Setting=Base2026.02 | 0 | 0 | 28 | 43 | 8 | 3 | 12 | 5 | 0 | 9 |