aiml-nsfw-benchmark

What problem does it solve?

The NSFW benchmarking template provides a standardized, reproducible framework to evaluate DistilBERT NSFW text classification across multiple explicit content categories, enabling researchers to measure recall, coverage, and robustness in a controlled setting.

Core Features & Use Cases

• Standardized NSFW categories (sexual_explicit, sexual_suggestive, harassment_sexual, grooming, sextortion) and language variants to test cross-language capabilities.
• Ready-to-use prompts and sample datasets with deterministic evaluation flows, plus guidance for extending categories and anchors.
• Use cases include model evaluation in AI safety research, benchmark-driven fine-tuning, and validation of prompt-variant experiments.

Quick Start

Run the included prompts and scripts to evaluate your NSFW classifier on the provided five-category benchmark.