Persona Consistency on Consistent-LLMs therapy domain
94Persona Consistency3-Step-PG
Evaluation Results
| Method | Links | |
|---|---|---|
| 3-Step-PGBackbone=Qwen, Evaluation methodology=Step-level2026.03 | 94 | |
| 3-Step-PGBackbone=Qwen, Evaluation methodology=Rollout2026.03 | 93.7 | |
| 2-Step-PGBackbone=Qwen, Evaluation methodology=Step-level2026.03 | 93.6 | |
| PPOBackbone=Qwen, Evaluation methodology=Step-level2026.03 | 91.1 | |
| 5-Step-PGBackbone=Qwen, Evaluation methodology=Step-level2026.03 | 89 | |
| GreedyPGBackbone=Qwen, Evaluation methodology=Step-level2026.03 | 88.2 | |
| BaseBackbone=Qwen, Evaluation methodology=Step-level2026.03 | 86.9 | |
| 5-Step-PGBackbone=Qwen, Evaluation methodology=Rollout2026.03 | 86.9 | |
| 2-Step-PGBackbone=Qwen, Evaluation methodology=Rollout2026.03 | 85.4 | |
| GreedyPGBackbone=Qwen, Evaluation methodology=Rollout2026.03 | 84 | |
| 4-Step-PGBackbone=Qwen, Evaluation methodology=Rollout2026.03 | 83.5 | |
| 4-Step-PGBackbone=Qwen, Evaluation methodology=Step-level2026.03 | 83 | |
| PGBackbone=Qwen, Evaluation methodology=Step-level2026.03 | 81.5 | |
| PGBackbone=Qwen, Evaluation methodology=Rollout2026.03 | 78.5 | |
| 3-Step-PGBackbone=Llama, Evaluation methodology=Step-level2026.03 | 78.4 | |
| 3-Step-PGBackbone=Llama, Evaluation methodology=Rollout2026.03 | 78 | |
| PPOBackbone=Qwen, Evaluation methodology=Rollout2026.03 | 75.3 | |
| 5-Step-PGBackbone=Llama, Evaluation methodology=Step-level2026.03 | 74 | |
| 2-Step-PGBackbone=Llama, Evaluation methodology=Step-level2026.03 | 73.4 | |
| 2-Step-PGBackbone=Llama, Evaluation methodology=Rollout2026.03 | 71.2 | |
| GreedyPGBackbone=Llama, Evaluation methodology=Step-level2026.03 | 71 | |
| 5-Step-PGBackbone=Llama, Evaluation methodology=Rollout2026.03 | 69 | |
| 4-Step-PGBackbone=Llama, Evaluation methodology=Step-level2026.03 | 68.7 | |
| GreedyPGBackbone=Llama, Evaluation methodology=Rollout2026.03 | 68.5 | |
| PPOBackbone=Llama, Evaluation methodology=Step-level2026.03 | 67.9 | |
| 4-Step-PGBackbone=Llama, Evaluation methodology=Rollout2026.03 | 67.3 | |
| PPOBackbone=Llama, Evaluation methodology=Rollout2026.03 | 62.5 | |
| PGBackbone=Llama, Evaluation methodology=Step-level2026.03 | 59.5 | |
| PGBackbone=Llama, Evaluation methodology=Rollout2026.03 | 58.6 | |
| BaseBackbone=Qwen, Evaluation methodology=Rollout2026.03 | 56.1 | |
| BaseBackbone=Llama, Evaluation methodology=Rollout2026.03 | 46.5 | |
| BaseBackbone=Llama, Evaluation methodology=Step-level2026.03 | 46.5 |