Text Quality Evaluation on C4 20 prompts (test)
3.2FluencyNo Watermark
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| No WatermarkBase Model=OPT-6.7B, Judge Model=GPT-4o, Watermark Strength (delta)=N/A2024.04 | 3.2 | 3.15 | 3.85 | 3 | |
| TBWBase Model=OPT-6.7B, Judge Model=GPT-4o, Watermark Strength (delta)=2.02024.04 | 3.1 | 3.05 | 3.9 | 3 | |
| SynthIDBase Model=OPT-6.7B, Judge Model=GPT-4o, Watermark Strength (delta)=2.02024.04 | 3 | 3.05 | 3.7 | 2.9 | |
| UnigramBase Model=OPT-6.7B, Judge Model=GPT-4o, Watermark Strength (delta)=2.02024.04 | 2.65 | 2.7 | 3.35 | 2.9 |