PyTorchCon to showcase warm-start Helion autotuning with shared caches for faster GPU kernel iteration

PyTorch · x · 2026-09-24

PyTorch announced that at PyTorch Conference North America (Oct 20-21, San Jose), Red Hat senior software engineer Alessandro Sangiorgi will present a poster, "Accelerating Helion Autotuning with Warm-Start and Shared Caches."

Helion is a PyTorch DSL for GPU kernels; the talk covers how warm starts let autotuning reuse cached configurations so developers iterate faster instead of repeating cold search runs.

Registration details: standard late tickets $999, academic tickets $249, 25% discount for teams of 10+.

Related event: PyTorchCon to Showcase Helion Autotuning Warm Starts(2 posts)→

Original post →

More from Infra

Infra channel →