Instruction Generalization on RoleBench instruction generalization
55.8GPT-4 Win RateRoleLLaMA-7B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| RoleLLaMA-7BBackbone=LLaMA-7B2023.10 | 55.8 | 52 | |
| Vicuna2023.10 | 32 | 23.4 | |
| Character.AI2023.10 | 31.4 | 30.2 | |
| Alpaca2023.10 | 16 | 20 | |
| ChatPLUG2023.10 | 3.8 | 16.4 |