Toxicity Steering on 15 prefix prompts length 50
71.2Toxicity AccuracyILRR
Evaluation Results
| Method | Links | |
|---|---|---|
| ILRRBase Model=LLaDA, alpha1=0.95, alpha2=1.0, Sequence Length=50, Normalized NFE=22026.01 | 71.2 | |
| ILRRBase Model=LLaDA, alpha1=0.8, alpha2=0.85, Sequence Length=50, Normalized NFE=22026.01 | 60.7 | |
| ILRRBase Model=MDLM, alpha=1.0, Sequence Length=50, Normalized NFE=42026.01 | 15.7 | |
| FKBase Model=LLaDA, Sequence Length=50, Normalized NFE=42026.01 | 9 | |
| PG-DLMBase Model=LLaDA, Sequence Length=50, Normalized NFE=42026.01 | 8.3 | |
| ILRRBase Model=MDLM, alpha=0.8, Sequence Length=50, Normalized NFE=42026.01 | 7.2 | |
| FKBase Model=MDLM, phi=1, Sequence Length=50, Normalized NFE=42026.01 | 3.8 | |
| best-of-nBase Model=LLaDA, Sequence Length=50, Normalized NFE=42026.01 | 2.4 | |
| best-of-nBase Model=MDLM, Sequence Length=50, Normalized NFE=42026.01 | 1.9 | |
| PG-DLMBase Model=MDLM, Sequence Length=50, Normalized NFE=42026.01 | 1.4 | |
| FKBase Model=MDLM, phi=4, Sequence Length=50, Normalized NFE=42026.01 | 0.8 |