TensorFlow is a machine learning framework developed by the Google Brain team as the successor to its internal DistBelief system and released as open source in November 2015. It represents computations on multidimensional arrays (tensors) as a dataflow graph, trains models with automatic differentiation, and runs on CPUs, GPUs, and TPUs. Starting with TensorFlow 2.0 in 2019, eager execution became the default and the high-level Keras API was integrated.
Its broad ecosystem, including TensorFlow Lite (now LiteRT) for mobile and embedded devices, TensorFlow.js for browsers, and TensorFlow Serving for server deployment, made it a common choice for production deployment. PyTorch has since become the mainstream framework in research and much of Google's own research has moved to JAX, but TensorFlow remains widely used in existing products and educational material.