PyTorch
PyTorch is an open-source machine-learning framework that lets developers build, train, and deploy deep-learning models with Pythonic ease. Released by Facebook AI Research in 2017, it offers dynamic computational graphs—operations run immediately, so debugging feels like standard Python. Core components include torch.Tensor for GPU-accelerated arrays, the autograd engine for automatic differentiation, and torch.nn for stacking layers into neural networks. The ecosystem spans data loaders, distributed training via torch.distributed, quantization for edge devices, torch.export with AOTInductor for ahead-of-time compilation and C++ deployment, and ExecuTorch for mobile and edge; TorchScript, the earlier export path, is now in maintenance mode. PyTorch backs breakthroughs in vision, NLP, and reinforcement learning, and serves production at Meta, Tesla, and OpenAI. A regular release cadence keeps CUDA, ROCm, and Apple Silicon (MPS) support current, with nightly builds in between, while a large ecosystem of community projects supplies pretrained models on Hugging Face and libraries like fastai and Lightning. By combining research flexibility with production tooling, PyTorch has become the go-to platform for modern AI.
Related terms
Related services: Agentic AI consulting.
Ready to put agentic AI to work?
Book a free 45-minute consultation. We'll map one real process worth automating with production-grade AI.