TransformerEngine icon indicating copy to clipboard operation
TransformerEngine copied to clipboard

Tests for distributed

Open pggPL opened this issue 1 year ago • 1 comments

Description

I am working on debug API. Before it can be merged, it needs to be tested. We need to ensure that all the new layers will work properly in the distributed case also. The tests present in the repo focus on testing gemm/comm overlapping - the tests I want to add focus more on checking numerical corectness of multiple configurations of TE layers. Moreover, current

Type of change

  • [ ] Documentation change (change only to the documentation, either a fix or a new content)
  • [ ] Bug fix (non-breaking change which fixes an issue)
  • [ ] New feature (non-breaking change which adds functionality)
  • [ ] Breaking change (fix or feature that would cause existing functionality to not work as expected)
  • [ ] Infra/Build change
  • [ ] Code refractor

Changes

  • added numerical tests for distributed training,

Checklist:

  • [x] I have read and followed the contributing guidelines
  • [x] The functionality is complete
  • [x] I have commented my code, particularly in hard-to-understand areas
  • [x] I have made corresponding changes to the documentation
  • [x] My changes generate no new warnings
  • [x] I have added tests that prove my fix is effective or that my feature works
  • [x] New and existing unit tests pass locally with my changes

pggPL avatar Sep 23 '24 09:09 pggPL

As mentioned by @ptrendx, we'll need to include these tests in one of the QA scripts (see qa) so that they are included in the CI pipelines. L1_pytorch_distributed_unittest is simplest, but we could also make a new L0 distributed test. If we go that route, it might be worth including the te.Sequential distributed tests as well.

timmoon10 avatar Sep 23 '24 17:09 timmoon10