Enable multi-device for some models#30207
Conversation
|
Hey @amyeroberts! We have added support for 10 models now. Can I get a review for this PR? |
amyeroberts
left a comment
There was a problem hiding this comment.
Amazing piece of work - thanks for adding this feature for all these models! ❤️
For the quality checks, running make fix-copies and pushing the changes should resolve this.
Could you update the issue to mark all of these models as done once merged in?
|
Thanks Amy! I ran |
|
@jackylee328 Ah yes - that's what Thanks for running tests for the new models too ❤️ Re the failing checks at the moment:
|
|
Finally passing now! I had to merge all the changes from main. |
|
Hmm I'm not sure why the torch tests are failing. I'm not able to reproduce the errors on my machines |
|
@jackylee328 Unfortunately sometimes our test suite will fail for reasons unrelated to the PR e.g. timeouts. If this happens, feel free to ping on the PR and I can restart the runs for you, without you needing to push lots of commits! Thank you for all the efforts adding this across our library - it's a mammoth addition! 🔥 |
|
Thank you @amyeroberts. I don't have permission to update the tracking issue, but I can make a list of the models that are now supported. |
|
@jla524 Ah, sorry, I didn't realise. No worries - I can update! |
* feat: multidevice for resnet * feat: yes! resnet * fix: compare all elements in tuple * feat: support for regnet * feat: support for convnextv2 * feat: support for bit * feat: support for cvt * feat: add support for focalnet * feat: support for yolos * feat: support for glpn * feat: support for imagegpt * feat: support for levit * feat: support for mgp_str * feat: support for mobilnet_v1 * feat: support for mobilnet_v2 * feat: support for mobilevit * feat: support for mobilevitv2 * feat: support for poolformer * fix: copies * fix: code quality check * update: upstream changes from main * fix: consistency check * feat: support for sam * feat: support for switchformer * feat: support for swin * feat: support for swinv2 * feat: support for timesformer * feat: suport for trocr * feat: support for upernet * fix: check copies * update: rerun CI * update: rerun again, maybe * update: one more rerun --------- Co-authored-by: Jacky Lee <[email protected]>
What does this PR do?
Fixes #29786 (issue)
Includes a fix for unit tests on Backbone models, where
base_output[0]andnew_output[0]are tuplesTested on a system with 2x RTX A4000
^ test skipped due to
CUDA error: misaligned addresswith PyTorch 2.0.0. which occurs when running on single GPU tooWho can review?
@amyeroberts