Build, train, and deploy machine learning models with an open-source platform
Writing a few lines of Python code lets you immediately verify the basic operations of a machine learning model, from numerical calculations to text processing. Originally created by researchers at Google Brain for neural network research, it has become a comprehensive toolkit for both researchers and developers. It provides the libraries and community resources needed for model building, training, and deployment, allowing you to start without complex environment setup.
An integrated ecosystem for research and development
TensorFlow is a platform that bundles tools, libraries, and community resources for machine learning research and application development. Developed by the Machine Intelligence team at Google Brain for neural network research, it is designed to be used in various domains rather than being limited to a specific field. Researchers can experiment with advanced techniques, while developers can build and deploy machine learning-based applications. This flexibility allows a single tool to support the workflow from the research stage to actual service deployment.
Stability of Python and C++ APIs
TensorFlow guarantees the stability of its Python and C++ APIs, while providing APIs for other languages without backward compatibility guarantees. The current release version supports GPU cards with CUDA enabled and is available in Ubuntu and Windows environments. DirectX and MacOS-metal environments are supported through Device Plugins. In environments without a GPU, you can install the smaller CPU-only package tensorflow-cpu. To update to the latest version, add the --upgrade flag to the pip install command.
Installation and first run
The default installation method is to run the pip install tensorflow command. Various installation options, such as enabling GPU support, using Docker containers, and building from source, can be found in the official installation guide. For testing, you can also use nightly binaries from the tf-nightly and tf-nightly-cpu packages on PyPI. Official builds are provided via PyPI or download links for various platforms, including Linux CPU/GPU, macOS, Windows CPU/GPU, Android, and Raspberry Pi. For platforms where official builds are not provided, you should check the community build table.
Version management and community participation
To apply bug or security vulnerability fixes to a specific version, you must switch to that version's branch (e.g., r2.8), apply the changes, and pass the tests. Afterward, you build and distribute the pip package from the source. GitHub Issues are used for tracking requests and bugs, while general questions and discussions are directed to the TensorFlow forum, and specific questions to Stack Overflow. This project follows common best practices for open-source software development, and participants must adhere to the code of conduct. Subscribing to the announce@tensorflow.org mailing list is recommended to receive security updates and release announcements.
License and support scope
This repository follows the Apache License 2.0, allowing commercial use, modification, and distribution. However, APIs for other languages do not have backward compatibility guarantees, so caution is required when developing in those languages. GPU support is limited to CUDA-based cards, while other graphics interfaces are supported through plugins. For platforms where official builds are not provided, you should check the community build table. Subscribing to the announce@tensorflow.org mailing list is recommended to receive security updates and release announcements.