Hardware-Aware Python Packages ~ PyTorch and...

Lightning Talk: Hardware-Aware Python Packages ~ PyTorch and WheelNext Grab the Wheel! - Jonathan Dekhtiar, NVIDIA & Eli Uriegas, Meta The PyTorch ecosystem thrives on innovation and a vibrant open-source community. PyTorch’s reach continues to evolve, fueled today by specialized hardware, variations within CPU architecture families, dedicated infrastructures, and diverse use cases. The Open-Source WheelNext (https://wheelnext.dev/) project joined forces with PyTorch to redesign and evolve how Python packaging work, allowing to better target compute environments with specialized hardware In this talk, we’ll describe how the PyTorch team - in partnership with NVIDIA, the WheelNext project and its numerous contributors - are developing the specifications for “Wheel Variants”. This format aims to smooth the PyTorch installation user experience, automatically detecting the AI accelerator or hardware available on the machine This talk aims to provide you an overview of the work progress, give you an insider look of the future changes we hope to introduce to the Python packaging standard and give you back the steering-Wheel of your installation experience and story. We invite every PyTorch practitioner and developer to test, provide feedback on this experimental release and help us build the future of Python packaging