How to turn off inlining / force materialization in TorchInductor during torch.compile?

torch.compile torch._inductor
1

Hi all,

I’m trying to reproduce the “Without inlining” ablation from the PyTorch 2.0 paper (the table that reports speedups for “All TorchInductor optimizations”, and then “Without inlining”, “Without fusion”, and “Without fusion & inlining”, etc.). My research goal is to turn off inlining (as well as other optimizations) in TorchInductor while keeping the rest of the pipeline intact, so I can attribute speedups precisely.

image
image571×283 54.4 KB

Questions

  1. Is there an official flag or env var to disable inlining (i.e., force materialization) in TorchInductor? If so, what is its name and recommended usage?
  2. How do I verify inlining is off?

Thanks!