BERT Small Multilingual Toxicity Classification

Supported Languages:

• Node.js
• Python

Statistics:

• 96.6% final train accuracy
• 94.6% final validation accuracy
• ≈35 supported languages

Dataset:

• 2.7M subset of https://huggingface.co/datasets/FredZhang7/toxi-text-3M

Setup:

1. Move variables.data-00000-of-00001 and variables.index to a folder named variables.
2. Move saved_model.pb to the same directory as variables.
3. Put those files under bert-small-multilingual.
4. Make sure you are in the same directory as bert-small-multilingual.
5. Run predict.js in the main branch of the repo.

Documented Training:

• https://huggingface.co/FredZhang7/one-for-all-toxicity-v3

License:

• Apache 2.0