Content policy on RoGuard
59.1AccuracyLlama-3.3-70B-Instruct-FP8
Evaluation Results
| Method | Links | |
|---|---|---|
| Llama-3.3-70B-Instruct-FP8N (#rows)=999, K (#candidate labels)=23, Decoding Strategy=greedy2026.05 | 59.1 | |
| gpt-oss-120BN (#rows)=999, K (#candidate labels)=23, Decoding Strategy=prefill-only logit scoring2026.05 | 55.5 |