pytorch
Tensors and Dynamic neural networks in Python with strong GPU acceleration
What is pytorch?
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Overview of pytorch
What it is
PyTorch is a Python package for tensor computation with GPU acceleration and deep neural networks built on a tape-based autograd system. The README describes it as deeply integrated into Python rather than a binding into a monolithic C++ framework. Its components include torch, torch.autograd, torch.nn, torch.multiprocessing, and torch.utils.
Key features
- Tensor computation similar to NumPy, with tensors that can live on CPU or GPU.
- Tape-based automatic differentiation through torch.autograd for differentiable tensor operations.
- torch.nn, a neural networks library integrated with autograd.
- torch.multiprocessing for Python multiprocessing with memory sharing of torch Tensors across processes.
- torch.utils, including DataLoader and other utility functions.
- Imperative execution: executed lines run immediately, and stack traces point to where code was defined.
- Extension options in Python or C/C++, with an extension API intended to require minimal boilerplate.
Use cases
The README lists two common uses: as a replacement for NumPy to use GPUs, and as a deep learning research platform. It also notes data loading and Hogwild training for torch.multiprocessing, and writing new neural network layers in Python or C/C++.
Getting started
The README does not provide a complete installation command. It states that Conda and pip wheel installation commands are on the project website, and it documents building from source. Source prerequisites include Python 3.11 or later, a compiler that fully supports C++20, Visual Studio or Visual Studio Build Tool on Windows, at least 10 GB of free disk space, and 30-60 minutes for the initial build. See the repository README for details.
License and pricing
The provided metadata lists the license as NOASSERTION and the pricing type as open_source. The README contains a License section, but the exact license terms are not stated in the provided text.
Based on the project's public README and repository metadata.