Harmful Conversation Detection on ProsocialDialog 652-conversation (test)
98PrecisionLlamaGuard
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LlamaGuardmodel_size=8B2026.01 | 98 | 50 | 66 | |
| Keyword2026.01 | 95 | 32 | 48 | |
| Lattice2026.01 | 90 | 93 | 91 | |
| NeMo2026.01 | 82 | 93 | 87 |
| Method | Links | |||
|---|---|---|---|---|
| LlamaGuardmodel_size=8B2026.01 | 98 | 50 | 66 | |
| Keyword2026.01 | 95 | 32 | 48 | |
| Lattice2026.01 | 90 | 93 | 91 | |
| NeMo2026.01 | 82 | 93 | 87 |