Safety IndexEvaluationsPolyGuardPrompts
PolyGuardPrompts
Adversarial robustness · Multilingual safetyCounts in the indexBuilt, not yet run
Whether the model refuses harmful prompts and answers benign ones across 17 languages.
- Metric
- % unsafe responses. Lower is better.
- Dataset
- huggingface.co/datasets/ToxicityPrompts/PolyGuardPrompts · test split subsample 1,700 · check card
- Grader
- PolyGuard-Qwen (paper's own openifier, run locally)
- Status
- Built, not yet run
- Replicability
- public
Results
Built and tested; the first full runs are pending.