license: mit
WildGuardMix Set with Classifier Predictions
This dataset contains a 90% split of the allenai/wildguardmix train subset, augmented with predictions from a fine-tuned ModernBERT-large classifier ensemble.
Contents
- Source: allenai/wildguardmix (wildguardtrain subset)
- Split: Annotation set (77,646 samples)
- Stratification: Stratified on prompt_harm_label, adversarial, and subcategory
Added Columns
harmful_probability: Model's predicted probability that the prompt is harmful (class 1)predicted_label: Binary prediction (0=unharmful, 1=harmful)label: Ground truth binary label (0=unharmful, 1=harmful)
Model Performance
The ensemble classifier achieved an F1 score of 95.8% on a separate test set which:
Intended Use
This dataset is designed for annotating difficult examples with natural language descriptions of prompt intent. The harmful_probability can be used to identify edge cases and ambiguous prompts for further analysis.
Terms and Conditions
By using this dataset, you agree to the terms and conditions specified by the original dataset provider (allenai/wildguardmix). Please refer to the original dataset's license and usage terms.