Skip to content

Suggestion: integration of einops test suite #146782

@arogozhnikov

Description

@arogozhnikov

Hi torch team,

Starting from einops 0.8.1, you can test torch against einops with:

# install numpy, einops, pytest and torch
python -m einops.tests.run_tests numpy torch

and I suggest having this in torch's CI.

There are a couple of motivations:

  1. einops tests actually reveal regressions in frameworks (happened several times, not in torch)
  2. it is hard within einops to test against more advanced features like torch.compile, because most of engineering/regressions happen on torch side. If tests fail within einops - 1) that's late, problem should be caught earlier 2) not much I can do. See this issue for a motivational example. Should simplify work on your side too.

Ready to answer questions / discuss concerns. Tests currently take ~12 seconds.

cc @seemethere @malfet @pytorch/pytorch-dev-infra @mruberry @ZainRizvi @jianyuh @nikitaved @pearu @walterddr @xwang233 @lezcano

Metadata

Metadata

Assignees

Labels

module: ciRelated to continuous integrationmodule: linear algebraIssues related to specialized linear algebra operations in PyTorch; includes matrix multiply matmulmodule: testsIssues related to tests (not the torch.testing module)triagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

Type

No type

Projects

No projects

Milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions