Inside Helion: Live Q&A with the Developers
Helion — a new Python-embedded domain-specific language (DSL) for wiring high-performance ML kernels — is built to compile down to Triton. Helion raises the level of abstraction for kernel authoring with PyTorch-centric syntax, making it dramatically easier to write correct, fast, and portable kernels. Its innovative autotuning system explores thousands of candidate Triton kernels to find the most performant implementation. Thus, automation handles complex, hardware-dependent details to ensure portability across different hardware architectures.
Now is your chance to meet the developers of Helion in a live Q&A!
Join us as we dive into:
- The design philosophy and architecture of Helion
- How autotuning delivers state-of-the-art performance across GPUs
- What’s next on the roadmap — and how you can get involved
Whether you’re just curious about Helion or already experimenting with it, come ask your questions, share ideas, and connect with the team making performant kernel authoring easier.
Bios:
- Jason Ansel: Research Scientist at Meta and the creator of PyTorch Compiler and Helion DSL
- Oguz Ulgen: Software Engineer at Meta and the creator of the PyTorch Compiler cache, working on Helion
- Will Feng: Software Engineer at Meta, working on PyTorch Compiler and Helion
- Jongsok Choi: Engineering Manager at Met, supporting the PyTorch Compiler Backend (Inductor) team and Helion