Introduction to Keras Core and its Transition to Keras 3
Keras Core is an exciting project within the world of deep learning, and it represents a significant evolution in the framework known as Keras. Initially, during its development stages from April 2023 to September 2023, Keras Core was the codename for a major update to the existing Keras library. It aimed at integrating multiple backend functionalities to provide a more versatile and powerful toolset for machine learning practitioners.
Background and Development
Keras Core was actively developed and tested publicly over several months. The project's goal was to enhance Keras by allowing it to support multiple backends efficiently. This improvement enables developers to switch between different computational backends more seamlessly, such as TensorFlow, Theano, and CNTK, thereby offering greater flexibility depending on the specific needs of a project.
Transition to Keras 3
With its development and public beta testing completed, Keras Core is transitioning to become what will be known as Keras 3. This transition signifies a new era for Keras, reflecting the advancements and updates made through Keras Core's development phase. Now, the development and updates related to Keras are carried out in the repository keras-team/keras.
Repository Changes
Previously, contributions were made to a separate project named keras-team/keras-core. However, with the shift to Keras 3, all future issues, pull requests, and contributions should now be directed to the keras-team/keras repository. The older keras-team/keras-core repository has been retired, streamlining development under one unified location.
Legacy tf.keras Codebase
As part of this organizational shift, the legacy codebase known under tf.keras is now housed separately. It has been relocated to a distinct repository at keras-team/tf-keras. This move preserves the old structure for those who might still rely on the traditional TensorFlow implementation of Keras, ensuring backward compatibility and ease of reference.
Preserving Contributions
For those developers who contributed to the Keras Core project during its codename phase, be assured that all contributions have been meticulously preserved. Your commitments have been reapplied in the keras-team/keras repository under your own authorship, maintaining a complete record of your efforts. This process has also been followed for previous contributors to keras-team/keras, integrating the new enhancements on top of the previous work.
Conclusion
The evolution of Keras Core into Keras 3 marks a pivotal moment in the machine learning framework's development, offering more versatility and a concentrated effort in furthering its capabilities. The move simplifies the development structure and encourages continued collaboration in the centralized keras-team/keras repository, ensuring that Keras evolves to meet the needs of modern machine learning challenges.