https://lalitm.com/tags/pytorch/Recent content in Pytorch on Lalit MagantiHugoenFri, 03 Jul 2026 00:23:00 +0100https://lalitm.com/post/perfetto-v57/Fri, 03 Jul 2026 00:23:00 +0100https://lalitm.com/post/perfetto-v57/<p>We just <a href="https://github.com/google/perfetto/releases/tag/v57.1">released Perfetto v57</a> and I wanted to share the new things I&rsquo;m most excited about. This is something I wanted to do for past releases but I just never quite got round to it. It&rsquo;s also something I plan on doing more of going forward: there might even be dedicated pieces if I think the feature deserves it!</p> <h2 id="pytorch-traces-reliably-work">PyTorch traces reliably work!</h2> <p>What I&rsquo;m most excited about in this release isn&rsquo;t a feature but a bugfix. If you used the <a href="https://docs.pytorch.org/tutorials/recipes/recipes/profiler_recipe.html">PyTorch profiler</a> and opened the resulting trace in Perfetto, there was a decent chance some of your events would just not show up; specifically this would happen when these events overlapped each other on a single track.</p> <p>Technically, PyTorch is in the wrong here. The Chrome Trace Event (JSON) format says duration events on a track have to nest and can&rsquo;t overlap; if you need overlap, you&rsquo;re supposed to use async events. <code>chrome://tracing</code> appears to handle them, but its rendering is actually buggy as soon as a trace has real overlaps; people just learned to live with it. So when a <a href="https://github.com/google/perfetto/issues/4280">bug</a> came in January about overlapping events being broken, I closed it as working-as-intended because I couldn&rsquo;t see any easy fix from our end.</p>