pytorch / pytorch/executorch

XNNPack configurations can be complicated to get right

Open
#8,884 14 comments 3 reactions 1 assignee View on GitHub

@mcr229 is already working on this.

Since Mar 3, 2025.

module: user experience module: xnnpack triaged
Dominant language
Python
Stars
5k
Forks
1.2k
Avg merge
2d 10h
Merged PRs (30d)
581

Description

It can be a little confusing to get XNNPack quantization + delegation configuration right. As seen in this issue for example the user to had to go through quite a bit of iteration to get to the right combination
1.) They had to figure out to add DuplicateDynamicQuantChainPass
2.) Then they had to do this to get delegation working https://github.com/pytorch/executorch/issues/8727#issuecomment-2694795413

I think for cases like 1) we should ideally not need the users to explicitly add this pass.
For cases like 2) it can be confusing to users whether to call XnnpackDynamicallyQuantizedPartitioner or XnnpackPartitioner etc. Not really sure what's the best solution for this but will leave it to @mcr229 @digantdesai

cc @digantdesai @mcr229 @cbilgin @mergennachin @byjlw

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.