The TensorFlow project announced the release of version 2.3.0, featuring new mechanisms for reducing input pipeline bottlenecks, Keras layers for pre-processing, and memory profiling.

TensorFlow developer advocate Josh Gordon outlined the highlights of the new release in a recent blog post. The package includes a new service API which can distribute input preprocessing to a cluster of worker machines, which can increase the throughput of data during training. Additionally, a snapshot API can persist the results of the preprocessing pipeline to disk, reducing the work required on subsequent training runs. An experimental Keras Preprocessing Layers API allows some preprocessing operations to be incorporated into the deep-learning models, simplifying deployment of the models. The TF Profiler includes new tools for memory profiling and Python tracing, to assist in debugging. Gordon commented on Twitter that TensorFlow 2.3 is a “solid, user-focused release.”

#tensorflow #google #deep learning #ai # ml & data engineering #news

TensorFlow 2.3 Features Pipeline Bottleneck Reduction and Improved
1.40 GEEK