Multi-label Content Safety Classification on Synthetic Dataset
0.768F1 ScoreMarginal Probability
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Marginal ProbabilityModel=LLaMA Guard 32025.11 | 0.768 | 0.906 | |
| Marginal ProbabilityModel=LLaMA 3.1 8B Instruct2025.11 | 0.738 | 0.934 | |
| Conditional ProbabilityModel=LLaMA 3.1 8B Instruct2025.11 | 0.704 | 0.876 | |
| Greedy GenerationModel=LLaMA Guard 32025.11 | 0.701 | — | |
| Conditional ProbabilityModel=LLaMA Guard 32025.11 | 0.701 | 0.777 | |
| Joint ProbabilityModel=LLaMA Guard 32025.11 | 0.7 | 0.777 | |
| Probability UncertaintyModel=LLaMA Guard 32025.11 | 0.698 | — | |
| Entropy UncertaintyModel=LLaMA Guard 32025.11 | 0.698 | — | |
| Greedy GenerationModel=LLaMA 3.1 8B Instruct2025.11 | 0.697 | — | |
| LogTokUModel=LLaMA Guard 32025.11 | 0.669 | — | |
| Marginal ProbabilityModel=LLaMA Guard 22025.11 | 0.658 | 0.824 | |
| Greedy GenerationModel=LLaMA Guard 22025.11 | 0.644 | — | |
| Conditional ProbabilityModel=LLaMA Guard 22025.11 | 0.644 | 0.756 | |
| Joint ProbabilityModel=LLaMA Guard 22025.11 | 0.644 | 0.754 | |
| Joint ProbabilityModel=LLaMA 3.1 8B Instruct2025.11 | 0.626 | 0.874 | |
| LogTokUModel=LLaMA Guard 22025.11 | 0.533 | — | |
| Probability UncertaintyModel=LLaMA Guard 22025.11 | 0.496 | — | |
| Entropy UncertaintyModel=LLaMA Guard 22025.11 | 0.496 | — | |
| LogTokUModel=LLaMA 3.1 8B Instruct2025.11 | 0.462 | — | |
| Probability UncertaintyModel=LLaMA 3.1 8B Instruct2025.11 | 0.453 | — | |
| Entropy UncertaintyModel=LLaMA 3.1 8B Instruct2025.11 | 0.453 | — |