ytu-ce-cosmos/guardrail-tr

Guardrail-TR Guardrail-TR (ytu-ce-cosmos/guardrail-tr) is a large-scale Turkish prompt-safety dataset for training and evaluating content-moderation / guardrail classifiers. Dataset contains ~405K single-turn user prompts with: a binary safe / unsafe label, and multi-label hazard categories (a prompt may carry more than one category). All released prompt text is Turkish. Rows that originated from English sources were processed, translated and, where relevant, culturally… See the full description on the dataset page: https://huggingface.co/datasets/ytu-ce-cosmos/guardrail-tr.