adding stride_order before convolution
Before submitting
- [ ] Was this discussed/approved via a Github issue? (no need for typos and docs improvements)
- [ ] Did you read the contributor guideline, Pull Request section?
- [ ] Did you make sure to update the docs?
- [ ] Did you write any new necessary tests?
What does this PR do?
Fixes # (issue).
PR review
Anyone in the community is free to review the PR once the tests have passed. If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
Did you have fun?
Make sure you had fun coding 🙃
numerical issue looks a bit annoying. Changing the torch reference implementation using channels last. I thought this is automatically done in cudnn (NOT VERIFIED!), but there's still places where torch convolution goes through native kernels.
@jjsjann123 what's your view on this? The CI seemed to pass earlier, but the PR is still a draft and thus not merged...
sorry about keeping this one floating. I haven't got any chance to revist its impact on vision models, so I wasn't confidently pushing for it.
I'll close it for now and revisit when I have a bit more bandwidth.