Bert-Multi-Label-Text-Classification

Why we included this project

Multi-label classification is one of those jobs where a solid starting point beats a blank page, and this repo provides exactly that. It takes pretrained BERT and XLNet checkpoints and runs them through a complete PyTorch training setup, with data preprocessing, a learning-rate scheduler, training monitors, and per-label AUC reporting all included. The Jigsaw toxic-comment dataset serves as the worked example, and the README walks you through downloading the tokenizer files, editing a single config, and running three CLI commands that go from raw CSV to a trained classifier. The codebase is compact enough to read end to end, so it is a practical template for adapting the approach to your own labels. One caveat: the pinned transformers 2.5.1 version is older, so expect some dependency work before it runs on a current stack.