Support : Leverage Accelerate for object detection/segmentation models

[Tanmaypatil123]

What does this PR do?

Adding support for multi-GPU training in 6 object detection models and segmentation models.

Fixes # #28309

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.
@NielsRogge

[NielsRogge]

Hi, thanks for your PR. It looks like there's an issue with the segmentation models.

[Tanmaypatil123]

@NielsRogge could you take a look at this PR again . I have made required changes for segmentation models.

[NielsRogge]

Thanks, LGTM.

[amyeroberts]

Thanks for working on this @Tanmaypatil123!

Just a comment on the imports

[amyeroberts]

This needs to be behind a protected import:

if is_accelerate_available():
    from accelerate import PartialState
    from accelerate.utils import reduce

[amyeroberts]

Thanks for iterating! Changes look good to me.

Before merging could you:

• Run the slow tests for these models to confirm everything still passes? e.g. RUN_SLOW=1 pytest tests/models/conditional_detr/
• Test at least one of these models in a single and multi-GPU environment with the loss calculation i.e. labels passed in?
• Resolve conflicts

[Tanmaypatil123]

hey @amyeroberts removed conflicts .but whenever I tried to run test cases by RUN_SLOW=1 pytest tests/models/conditional_detr/ I am getting following error :

[amyeroberts]

@Tanmaypatil123 - Not sure exactly what's causing this. The function is_g2p_en_available was added in #23439. However, we haven't seen this import error in our CI runs which validate slow tests and I'm able to run RUN_SLOW=1 pytest tests/models/conditional_detr/ locally on main.

In a python session, are you about to run:

from transformers.utils import is_g2p_en_available

?

Could you try rebasing on main to make sure all updates/fixes are included in this branch?

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.

[ArthurZucker]

The issue is related to evaluate I think: huggingface/peft#1351

[ArthurZucker]

cc @Tanmaypatil123 let's rebase this and I think we could merge this!

[Tanmaypatil123]

@ArthurZucker Changes are approved but there is issue with segmentation model; test cases fails for it and i could not figure out what the problem is.

[NielsRogge]

@Tanmaypatil123 you just need to fix the upstream conflicts, like so:

git remote add upstream https://github.com/huggingface/transformers.git
git fetch upstream
git merge upstream/main

[Tanmaypatil123]

@ArthurZucker I have resolved the conflicts

[amyeroberts]

@Tanmaypatil123 Great!

For the failing tests, there's two which look like they might be related to this PR which will needed to be addressed before merge:

FAILED tests/models/yolos/test_modeling_yolos.py::YolosModelTest::test_pipeline_image_feature_extraction - TypeError: forward() got an unexpected keyword argument 'pixel_mask'
FAILED tests/models/yolos/test_modeling_yolos.py::YolosModelTest::test_pipeline_object_detection - TypeError: forward() got an unexpected keyword argument 'pixel_mask'

For the build documentation tests, there was a recent fix merged into main. Rebasing to include this should resolve.

[Tanmaypatil123]

@Tanmaypatil123 Great!

For the failing tests, there's two which look like they might be related to this PR which will needed to be addressed before merge:

FAILED tests/models/yolos/test_modeling_yolos.py::YolosModelTest::test_pipeline_image_feature_extraction - TypeError: forward() got an unexpected keyword argument 'pixel_mask'
FAILED tests/models/yolos/test_modeling_yolos.py::YolosModelTest::test_pipeline_object_detection - TypeError: forward() got an unexpected keyword argument 'pixel_mask'

For the build documentation tests, there was a recent fix merged into main. Rebasing to include this should resolve.

Don't know how that test case is failing. I didn't make any changes to the parameters that are provided to YolosForObjectDetection. should we add parameters or is there any change in the test case?

[amyeroberts]

@Tanmaypatil123 Indeed. It's quite weird these tests are failing as they don't seem related to this PR. They aren't however failing on any other PRs. What I think is happening is that some tests weren't collected by the test fetched (cc @ydshieh for reference) when the image feature extraction pipeline was added. And now, because this PR touches yolos they're being run now. I'm going to look into it and let you know asap.

[Tanmaypatil123]

@amyeroberts Done 😀 Thanks for the help. Huggingface has the best open source maintainers.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[amyeroberts]

@Tanmaypatil123 Thanks for this great contribution and patience with our failing CI tests!