Role-playing on RoleLLM
94.46GPT-4o Win RateFull System Prompt
Evaluation Results
| Method | Links | |
|---|---|---|
| Full System PromptBackbone=Qwen3-4B2025.11 | 94.46 | |
| [BE] TokenBackbone=Qwen3-4B, Tokens=12025.11 | 92.93 | |
| Full System PromptBackbone=Llama-3.1-8B2025.11 | 69.52 | |
| [BE] TokenBackbone=Llama-3.2-3B, Tokens=12025.11 | 67.54 | |
| Full System PromptBackbone=Llama-3.2-3B2025.11 | 65.62 | |
| Soft PromptBackbone=Qwen3-4B, Tokens=12025.11 | 65.32 | |
| [BE] TokenBackbone=Llama-3.1-8B, Tokens=12025.11 | 64.06 | |
| No System PromptBackbone=Llama-3.1-8B2025.11 | 60.97 | |
| Memory TokenBackbone=Qwen3-4B2025.11 | 51.12 | |
| Full System PromptBackbone=Llama-3.2-1B2025.11 | 47.26 | |
| Soft PromptBackbone=Llama-3.1-8B, Tokens=12025.11 | 46.52 | |
| [BE] TokenBackbone=Llama-3.2-1B, Tokens=12025.11 | 45.98 | |
| Soft Prompt (4 Tokens)Backbone=Qwen3-4B2025.11 | 45.53 | |
| No System PromptBackbone=Llama-3.2-3B2025.11 | 41.68 | |
| Soft Prompt (16 Tokens)Backbone=Qwen3-4B2025.11 | 40.22 | |
| No System PromptBackbone=Qwen3-4B2025.11 | 38.78 | |
| Soft PromptBackbone=Llama-3.2-3B, Tokens=12025.11 | 36.74 | |
| Soft Prompt (4 Tokens)Backbone=Llama-3.1-8B2025.11 | 34.46 | |
| Soft PromptBackbone=Llama-3.2-1B, Tokens=12025.11 | 31.57 | |
| Soft Prompt (4 Tokens)Backbone=Llama-3.2-3B2025.11 | 29.45 | |
| Soft Prompt (4 Tokens)Backbone=Llama-3.2-1B2025.11 | 28.57 | |
| Soft Prompt (16 Tokens)Backbone=Llama-3.2-3B2025.11 | 28.01 | |
| Soft Prompt (16 Tokens)Backbone=Llama-3.1-8B2025.11 | 27.95 | |
| Memory TokenBackbone=Llama-3.1-8B2025.11 | 22.13 | |
| Memory TokenBackbone=Llama-3.2-3B2025.11 | 18.82 | |
| No System PromptBackbone=Llama-3.2-1B2025.11 | 17.57 | |
| Soft Prompt (16 Tokens)Backbone=Llama-3.2-1B2025.11 | 11.08 | |
| Memory TokenBackbone=Llama-3.2-1B2025.11 | 0 |