Instruction Generalization on RoleBench Chinese instruction generalization 1.0
53.7ROUGE-L (CUS)RoleGPT
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| RoleGPT2023.10 | 53.7 | 57.5 | 24.8 | 45.3 | |
| RoleGLM2023.10 | 50.5 | 52.6 | 34.1 | 45.7 | |
| Character.AI2023.10 | 42 | 55.8 | 28.7 | 42.2 | |
| Yi-6B-Chatparameters=6B2023.10 | 40.6 | 56.5 | 26.9 | 41.3 | |
| ChatGLM2base_model=ChatGLM22023.10 | 39.4 | 50.6 | 31 | 40.3 | |
| ChatPLUG2023.10 | 38.9 | 61.4 | 31 | 43.8 | |
| ChatGLM2-scriptbase_model=ChatGLM2, variant=script-based2023.10 | 14 | 30.7 | 9.1 | 17.9 |