Skip to content

Conversation

@mattdangerw
Copy link
Member

TPU testing would be great, but we should not do it quite like this.

  • These are not run in any form of CI right now. Do they run at all? I don't think anyone has tried these for many months.
  • They basically replicate the simplest tests from the main fixture inside a TPU strategy scope. We would be much better off handling this inside some common test setup, rather than replicating test code.
  • They would not work multi-backend.
  • They do not test presets, tasks, backprop, and a lot of the actual important functionality for our pretrained model offering.

Let's remove for now as I consolidate our preset testing code. We can bring this back later when we have a plan to run these in a more sustainable way for our test suite.

TPU testing would be great, but we should not do it quite like this.

- These are not run in any form of CI right now. Do they run at all?
  I don't think anyone has tried these for many months.
- They basically replicate the simplest tests from the main fixture
  inside a TPU strategy scope. We would be much better off handling
  this inside some common test setup, rather than replicating test
  code.
- They would not work multi-backend.
- They do not test presets, tasks, backprop, and a lot of the
  actual important functionality for our pretrained model offering.

Let's remove for now as I consolidate our preset testing code. We
can bring this back later when we have a plan to run these in a more
sustainable way for our test suite.
@mattdangerw mattdangerw requested a review from ianstenbit October 5, 2023 19:57
@mattdangerw mattdangerw merged commit f0ab133 into keras-team:master Oct 6, 2023
mattdangerw added a commit that referenced this pull request Nov 7, 2023
TPU testing would be great, but we should not do it quite like this.

- These are not run in any form of CI right now. Do they run at all?
  I don't think anyone has tried these for many months.
- They basically replicate the simplest tests from the main fixture
  inside a TPU strategy scope. We would be much better off handling
  this inside some common test setup, rather than replicating test
  code.
- They would not work multi-backend.
- They do not test presets, tasks, backprop, and a lot of the
  actual important functionality for our pretrained model offering.

Let's remove for now as I consolidate our preset testing code. We
can bring this back later when we have a plan to run these in a more
sustainable way for our test suite.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants