Dialogue Safety Evaluation on Safety Bench Unit Tests
30Safe RateReddit 2.7B
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Reddit 2.7BParameter count=2.7B2022.05 | 30 | 26.1 | 45 | 43.9 | |
| OPT-175BParameter count=175B2022.05 | 3.3 | 26.1 | 56.7 | 28.3 | |
| BlenderBot 1Training=Fine-tuned on curated dialogue datasets2022.05 | 2.8 | 15 | 25 | 19.4 | |
| R2C2 BlenderBotTraining=Fine-tuned on curated dialogue datasets2022.05 | 2.2 | 13.3 | 28.9 | 22.2 |