Sentiment Control on IMDb PosToNeg (test)
63.1Success RatePREADD-S
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| PREADD-Sbase_model=OPT-6.7B, alpha=2, mode=static prefix2023.07 | 63.1 | 68.4 | 25.3 | |
| FUDGEbase_model=OPT-6.7B, sampling=top-k, discriminator=trained to predict sentiment2023.07 | 53.2 | 25.1 | 31.1 | |
| POSPROMPTbase_model=OPT-6.7B, prompt_type=specific sentiment prompting2023.07 | 30.7 | 53.5 | 29.8 | |
| Gbase_model=OPT-6.7B, description=base model2023.07 | 16.8 | 51.3 | 30.6 |