Character Consistency on CharacterEval
2.58KEPP
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| PPModels=GPT-4o2026.03 | 2.58 | 3.02 | 2.99 | 2.83 | 2.91 | 2.87 | |
| PPModels=LLaMA-3-8B-Instruct2026.03 | 2.53 | 2.72 | 2.44 | 2.88 | 2.53 | 2.62 | |
| Qwen-7B+PCLBackbone=Qwen-7B, Evaluation Mode=PCL2025.03 | 2.431 | 3.051 | 2.813 | 2.144 | 2.641 | 2.616 | |
| PDDModels=LLaMA-3-8B-Instruct2026.03 | 2.39 | 2.68 | 3.03 | 3 | 2.96 | 2.81 | |
| GPT-4+PCL*Backbone=GPT-4, Evaluation Mode=PCL* (Chain of Persona Prompting)2025.03 | 2.356 | 2.983 | 3.026 | 2.627 | 2.934 | 2.785 | |
| Qwen-7B+SFT+ICLBackbone=Qwen-7B, Evaluation Mode=SFT+ICL2025.03 | 2.336 | 2.945 | 2.671 | 2.225 | 2.582 | 2.551 | |
| Baichuan2-7B+PCLBackbone=Baichuan2-7B, Evaluation Mode=PCL2025.03 | 2.268 | 3.125 | 3.131 | 2.454 | 3.018 | 2.799 | |
| OPADModels=Qwen2.5-7B-Instruct2026.03 | 2.26 | 2.82 | 2.62 | 3.06 | 2.31 | 2.61 | |
| GPT-4+ICLBackbone=GPT-4, Evaluation Mode=ICL2025.03 | 2.25 | 2.855 | 2.785 | 2.721 | 2.873 | 2.697 | |
| PPModels=Qwen2.5-7B-Instruct2026.03 | 2.25 | 2.96 | 2.99 | 3.03 | 2.94 | 2.83 | |
| PDDModels=Qwen2.5-7B-Instruct2026.03 | 2.25 | 2.93 | 2.99 | 3.08 | 3.01 | 2.85 | |
| ICLModels=LLaMA-3-8B-Instruct2026.03 | 2.16 | 2.73 | 2.92 | 3.04 | 2.89 | 2.75 | |
| OPADModels=LLaMA-3-8B-Instruct2026.03 | 2.13 | 2.62 | 2.57 | 2.94 | 2.71 | 2.59 | |
| Baichuan2-7B+SFT+ICLBackbone=Baichuan2-7B, Evaluation Mode=SFT+ICL2025.03 | 2.115 | 3.067 | 3.05 | 2.511 | 2.967 | 2.742 | |
| SPModels=LLaMA-3-8B-Instruct2026.03 | 2.08 | 2.78 | 2.98 | 3.13 | 2.08 | 2.61 | |
| SPModels=Qwen2.5-7B-Instruct2026.03 | 2.07 | 3 | 3.23 | 3.09 | 2.38 | 2.75 | |
| GPT-3.5+PCL*Backbone=GPT-3.5, Evaluation Mode=PCL* (Chain of Persona Prompting)2025.03 | 2.037 | 2.67 | 2.468 | 1.856 | 2.372 | 2.28 | |
| Qwen-7B+ICLBackbone=Qwen-7B, Evaluation Mode=ICL2025.03 | 1.956 | 2.728 | 2.633 | 2.605 | 2.78 | 2.54 | |
| CharacterGLM-6B+PCLBackbone=CharacterGLM-6B, Evaluation Mode=PCL2025.03 | 1.934 | 2.543 | 2.278 | 1.485 | 2.138 | 2.076 | |
| ICLModels=Qwen2.5-7B-Instruct2026.03 | 1.92 | 2.79 | 2.92 | 2.72 | 2.42 | 2.55 | |
| GPT-3.5+ICLBackbone=GPT-3.5, Evaluation Mode=ICL2025.03 | 1.846 | 2.472 | 2.121 | 2.051 | 2.287 | 2.155 | |
| Baichuan2-7B+ICLBackbone=Baichuan2-7B, Evaluation Mode=ICL2025.03 | 1.809 | 2.853 | 2.919 | 2.84 | 3.081 | 2.7 | |
| CharacterGLM-6B+SFT+ICLBackbone=CharacterGLM-6B, Evaluation Mode=SFT+ICL2025.03 | 1.613 | 2.366 | 2.147 | 1.504 | 2.112 | 1.948 | |
| PPModels=Deepseek-R12026.03 | 1.42 | 3.07 | 2.83 | 1.93 | 2.89 | 2.43 | |
| CharacterGLM-6B+ICLBackbone=CharacterGLM-6B, Evaluation Mode=ICL2025.03 | 1.417 | 2.287 | 1.998 | 1.518 | 2.075 | 1.805 |