PyTorch at NVIDIA GTC 2026: Join Us in San Jose!
PyTorch's new Helion kernel framework and on-device ExecuTorch demos headline Meta's major presence at the 2026 conference.
Meta and the PyTorch Foundation are gearing up for a significant showcase at NVIDIA GTC 2026 in San Jose, with a packed schedule of demos, talks, and hands-on labs centered on the open-source framework's evolution. The centerpiece at Booth #338 will be Helion, a new PyTorch-native kernel authoring framework designed to let developers write and autotune custom kernels for optimal performance on NVIDIA GPUs. Alongside this, a live demo will feature ExecuTorch, PyTorch's deployment toolkit, running NVIDIA's Parakeet speech-to-text model for high-performance, Python-free, on-device inference, illustrating a complete edge AI pipeline.
Beyond the booth, the technical agenda is robust. A featured talk by PyTorch core maintainer Alban Desmaison, titled 'From Kernels to Clusters: How PyTorch Powers High-Performance AI,' will outline the framework's roadmap and its role in distributed workloads. Multiple sessions with experts from Meta, NVIDIA, and AWS will cover topics from turbocharging LLM inference and fault-tolerant training to hands-on labs for model optimization. Kicking off the week, a dedicated Helion Hackathon in San Francisco on March 14 offers developers a deep dive into GPU kernel programming with direct access to Meta and NVIDIA engineers.
- Showcases Helion, a new PyTorch-native framework for writing and autotuning custom GPU kernels.
- Live demo of ExecuTorch deploying NVIDIA's Parakeet model for Python-free, on-device speech-to-text inference.
- Features a keynote by core maintainer Alban Desmaison and a pre-conference Helion Hackathon on March 14.
Why It Matters
Highlights PyTorch's push for deeper hardware integration and efficient edge deployment, critical for next-gen AI performance.