https://duoan.github.io/tags/torch.compile/ Recent content in Torch.compile on Duo's Tech Blog https://duoan.github.io/images/papermod-cover.png https://duoan.github.io/images/papermod-cover.png Hugo -- 0.153.1 en-us Sun, 31 May 2026 00:00:00 +0000 https://duoan.github.io/posts/torch-compile-from-bytecode-to-triton/ Sun, 31 May 2026 00:00:00 +0000 https://duoan.github.io/posts/torch-compile-from-bytecode-to-triton/ <h1 id="torchcompile-the-mental-model-that-actually-matters">torch.compile: The Mental Model That Actually Matters</h1> <p>Most writeups of <code>torch.compile</code> are either a flag cheat-sheet or a file-by-file museum tour. Neither helps when a training step is only 1.2× faster and you do not know whether to blame graph breaks, recompiles, or Inductor.</p> <p>The useful model is simpler: <strong>compile is specialization under recorded assumptions</strong>. Dynamo captures a region, Inductor emits kernels tuned to that region, and guards decide whether the specialization still applies. Everything else — FX, AOTAutograd, Triton — is machinery in service of that contract.</p>