A Python library providing GPU-accelerated tensor computation and a dynamic neural network framework built on a tape-based autograd system. It is designed for researchers and developers who require flexibility and speed for deep learning model training and scientific computation.
Project overview
A Python library providing GPU-accelerated tensor computation and a dynamic neural network framework built on a tape-based autograd system. It is designed for researchers and developers who require flexibility and speed for deep learning model training and scientific computation.
Project type
Model Development · Infrastructure
Deployment
Refer to project documentation
License
License pending
Best for
Developers, researchers, and AI engineers who work with Python and require a library for model-training-customization with dynamic network behaviors.
Key capabilities
Provides Tensors that can live either on the CPU or the GPU and accelerates scientific computation.
Uses reverse-mode auto-differentiation to allow changing the way your network behaves arbitrarily with zero lag or overhead.
A neural networks library deeply integrated with autograd designed for maximum flexibility.
Creates serializable and optimizable models from PyTorch code.
Provides a convenient extension API to write new neural network layers in C/C++ with minimal boilerplate.
Limitations and risks
While GPU support is optional, building from source requires at least 10 GB of free disk space.
Building the library from source is rated as medium difficulty due to the need for C++ compilers and an extended build time.
Getting started
Install binaries via Conda or pip wheels, import the torch module, and build a neural network. Building from source requires C++ compilers and 30-60 minutes.
Evidence and sources
GitHub project description: Tensors and Dynamic neural networks in Python with strong GPU acceleration
README: PyTorch is a Python package that provides two high-level features: - Tensor computation (like NumPy) with strong GPU acceleration - Deep neural networks built on a tape-based auto…
README: PyTorch is a library that consists of the following components:
README: we use a technique called reverse-mode auto-differentiation, which allows you to change the way your network behaves arbitrarily with zero lag or overhead.
README: PyTorch provides Tensors that can live either on the CPU or the GPU and accelerates the computation by a huge amount.