Separius/BERT-keras
A Keras implementation of Google's BERT and OpenAI's Transformer for NLP tasks with pretrained model loading and fine-tuning support.

Velocity · 7d
+0.3
★ / day
Trend
→steady
star history
Provides a Keras-based implementation of BERT (Bidirectional Encoder Representations from Transformers) and OpenAI’s Transformer language model. The library enables loading pretrained weights and includes a fine-tuning API for downstream NLP tasks. It supports both single and double sentence tasks with masking, includes text encoding utilities such as sentencepiece and BPE, and offers TPU support for both training and inference.