bojone/bert4keras
A lightweight Keras reimplementation of BERT and transformer models with support for fine-tuning and pretraining.
Not currently ranked — collecting fresh signals.
star history
This project provides a clean, lightweight implementation of transformer models (BERT, RoBERTa, ALBERT) built on Keras. It supports loading pretrained weights for fine-tuning, implements necessary attention masks for language modeling and seq2seq tasks, and includes complete pretraining code supporting TPU and multi-GPU setups. The library is compatible with both Keras and tf.keras backends.
Frequently asked
- What is bojone/bert4keras?
- A lightweight Keras reimplementation of BERT and transformer models with support for fine-tuning and pretraining.
- Is bert4keras open source?
- Yes — bojone/bert4keras is open source, released under the Apache-2.0 license.
- What language is bert4keras written in?
- bojone/bert4keras is primarily written in Python.
- How popular is bert4keras?
- bojone/bert4keras has 5.4k stars on GitHub.
- Where can I find bert4keras?
- bojone/bert4keras is on GitHub at https://github.com/bojone/bert4keras.