Microsoft is now officially in charge of maintaining the Windows version of PyTorch, a popular open-source machine learning library created by Facebook.
Microsoft’s decision to take a bigger role in maintaining PyTorch for Windows is part of an effort to improve library performance on Windows 10 PCs and Windows Subsystem for Linux (WSL), which developers can use to run Linux distributions on Windows 10.
Facebook’s open source PyTorch in 2018, about a year after its launch to help developers build cutting-edge AI models.
Facebook has used PyTorch and the Caffe2 for Translate deep learning framework, an AI tool that powers translations of the 48 most widely used languages on Facebook.
PyTorch, as the name implies, is a package for the popular Python programming language. It helps developers use machine learning Python packages, like NumPy, and it helps with GPU-accelerated calculations for heavy data science tasks. PyTorch has also been one of the fastest growing projects on Microsoft’s GitHub.
According to PyTorch maintainers, the reason for the transfer is that support for PyTorch in Windows 10 has lagged behind that provided by Linux and macOS, despite Windows being the primary use of OS developers, according to the latest Stack Overflow developer survey.
“Lack of test coverage resulted in unexpected problems popping up from time to time. Some of the main tutorials, intended for new users to learn and adopt PyTorch, would not run,” Facebook and Microsoft engineers explain in a blog. set.
“The installation experience was not that easy either, with the lack of official PyPI support for PyTorch on Windows. Finally, some of PyTorch’s features were simply not available on the Windows platform, such as the TorchAudio domain library and support of distributed training.
“To help ease this pain, Microsoft is pleased to bring its expertise on Windows and put PyTorch on Windows in its best possible state.”
Moving PyTorch for Windows to Microsoft is related to the Redmond company’s efforts to improve WSL performance in Windows 10, which currently has pre-support for GPU-accelerated machine learning (ML) training.
WSL users have been demanding better GPU computing support to speed up ML training times, and the preview opens the door for developers and data scientists to use Nvidia’s CUDA platform to speed up training.
Microsoft and Nvidia last month released a preview of CUDA for WSL 2 with PyTorch support through new Windows graphics drivers that enabled CUDA in WSL 2.
As PyTorch maintainers point out, the preview gives developers the flexibility to work with multiple Python frameworks and packages that depend on Nvidia CUDA but are only compatible with Linux.
Preview means that “WSL clients using preview can run native Linux-based PyTorch applications on Windows without modifying them without the need for a traditional virtual machine or dual boot configuration.”
