appvision-ai/fast-bert

Provides fine-tuning pipelines for BERT, RoBERTa, XLNet, and DistilBERT with built-in support for learning rate discovery and the LAMB optimizer for accelerated training. Abstracts away PyTorch complexity through a DataBunch API that handles tokenization, data loading, and multi-GPU orchestration, while supporting both multi-class and multi-label text classification workflows.

1,920 stars. No commits in the last 6 months.