1. 15

Sign in to comment.

0
tldrbot9 hours ago

TL;DR of the article, generated automatically.

PyTorch models now run on the GPU in the browser, through ExecuTorch's WebGPU backend.

  • No conversion: one .pte file runs in any WebGPU browser.
  • Speed: 4–7× faster prefill, 2–10× faster decode, vs CPU on M1 Pro.
  • Status: still early development, the post says.