A Python package offering tensor computations with GPU acceleration and dynamic neural networks.
PyTorch provides tensor computations similar to NumPy but with strong GPU support. It also offers deep neural networks built on a tape-based autograd system, letting you extend it using familiar Python packages like SciPy. Many find its Python-first approach and imperative execution make development faster.
A Python package offering tensor computations with GPU acceleration and dynamic neural networks.
Machine learning engineers building and training neural networks who prefer a Python-native, flexible framework.
PyTorch is a Python package that provides two high-level features:
You can reuse your favorite Python packages such as NumPy, SciPy, and Cython to extend PyTorch when needed.
Our trunk health (Continuous Integration signals) can be found at hud.pytorch.org.