aiml-toxigen-benchmark

What problem does it solve?

This Skill provides a ready-to-use benchmark setup for evaluating implicit hate recall using ToxiGen HateBERT, streamlining the construction and validation of toxicity detectors within AI safety research.

Core Features & Use Cases

• Preconfigured benchmark: Includes a dataset with target groups and coded, implicit samples to test recall, plus validation scripts to enforce data quality constraints.
• Automated evaluation: Executes a text-classification model to score implicit hate for each sample and computes overall recall.
• Use Case: A researcher wants to validate that a new model's implicit-hate detection remains robust across diverse groups before deployment.

Quick Start

Run the toxigen benchmark against the provided dataset to assess how well a model detects implicit hate.