PyTorch Helion Hackathon

Join us to push the boundaries of GPU performance with Helion — PyTorch's new kernel authoring language. Alongside fellow kernel hackers and researchers, you'll get hands-on with the new Python-embedded DSL that compiles to Triton with innovative built-in autotuning — delivering higher performance in fewer lines of code. Performant GPU programming has never been this accessible.

Helion is the latest innovation from the PyTorch team. Just as PyTorch revolutionized deep learning with its flexible Pythonic design, Helion brings that same design philosophy to kernel development with its intuitive programming model: PyTorch with Tiles. If you know PyTorch, you already know most of Helion.

In addition, you will get a preview of NVIDIA's CompileIQ, which enables bespoke optimizations and code generation for higher performance.

Learn directly from PyTorch, Meta, and NVIDIA experts and compete to write the fastest Helion kernels on the GPU Mode leaderboard. Showcase your skills, help shape the future of the language with your feedback, and win some cool prizes!

Winning prizes include:

• NVIDIA DGX Spark, RTX 5090, and RTX 5080 GPUs

• GTC exhibit passes to demo your project at the Meta booth during NVIDIA GTC

• Ray-Ban Meta glasses

The judges will be:

• Jana van Greunen, Director of PyTorch Engineering, Meta

• Luis Ceze, Vice President of AI Systems Software, NVIDIA

• Ujval Kapasi, Vice President of AI & HPC Frameworks and Libraries, NVIDIA

• Horace He, Member of Technical Staff, Thinking Machines Lab

This hackathon is organized in partnership with Meta, PyTorch, GPU Mode, NVIDIA, and Nebius.

Note: This event is a fully in-person event in San Francisco, CA. Space is limited. Maximum team size is four. All attendees must submit an application for approval and be approved to attend.