Accelerated linear algebra: TensorFlow now supports accelerated linear algebra operations using the Eigen library, providing faster performance for matrix operations on CPUs.
Improved GPU support: TensorFlow has improved GPU support, with optimized kernels for common operations and better memory management for GPUs.
Eager execution: TensorFlow now supports eager execution, allowing for immediate computation of operations without needing to build a computation graph.
TensorFlow Lite: TensorFlow now has a lightweight version called TensorFlow Lite, designed for mobile and embedded devices.
Improved TensorFlow Serving: TensorFlow Serving now supports model versioning, allowing for easier management and deployment of models in production environments.
Improved TensorFlow Hub: TensorFlow Hub now provides a repository of pre-trained models and modules, making it easier to use pre-trained models in your own projects.
Improved TensorFlow.js: TensorFlow.js now supports training models directly in the browser, making it easier to build and deploy machine learning models in web applications.
Improved support for Keras: TensorFlow now has improved support for the Keras API, making it easier to build and train deep learning models using the high-level Keras API.