| --- |
| license: mit |
| task_categories: |
| - text-classification |
| language: |
| - en |
| tags: |
| - safety |
| - classification |
| size_categories: |
| - 10K<n<100K |
| --- |
| |
| ## Litesafe |
|
|
| Litesafe is a dataset and model pair for classifying safe vs unsafe text. It primaraly targets self harm, harm to others, strong sexual topics, and racism as bad things. This dataset was created synthetically with Deepseek v4 Flash making the text examples according to the classification, and extremeness and classification being chose randomly. |
| For the classifier, there are different levels of extremeness for each example. Extremeness is how strict the model is, with extremeness of 1 being basic curses and sometimes even more mild insults can get blocked, and extremeness of 4 allows curses and more insults but still blocks anything serious. |
|
|
| For classification in each example, 0 means safe and 1 means unsafe. |
