PyTorch

PyTorch is an open-source machine learning library for Python, developed primarily by Facebook AI Research (FAIR). It was released in 2016 as an alternative to the Lua-based Torch framework and has since become one of the most widely used tools for deep learning research and production.

Core concepts include tensors with GPU acceleration, automatic differentiation via autograd, and a dynamic computation graph that is built as operations are executed (define-by-run). The Pythonic interface facilitates rapid prototyping, while a C++ backend (libtorch) supports deployment in performance-critical environments.

Key components and ecosystem include TorchVision, TorchText, and Torchaudio, which provide domain-specific data and models. TorchScript enables serializing models for production, while TorchServe offers model serving. PyTorch Mobile targets deployment on iOS and Android. ONNX support enables interoperability with other frameworks and runtimes.

Usage and community: PyTorch is widely adopted in academia and industry for research in natural language processing, computer vision, and related fields. High-level frameworks such as PyTorch Lightning and fastai build on top of PyTorch. It competes with TensorFlow and benefits from a large and active contributor base.

License and development: PyTorch is open-source software released under the BSD license. It is maintained by Meta AI and a broad community of contributors, with builds available for Windows, macOS, and Linux.