Skip to content

Conversation

@andreeahedes
Copy link
Contributor

What does this PR do?

Fixes # (issue)
Add in no_split_modules method for PegasusX model to allow disk/cpu offloading and multi-GPU parallelism.

Before submitting

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.

@amyeroberts
Copy link
Contributor

amyeroberts commented Sep 4, 2023

@andreeahedes Thanks for adding this! Have you run the accelerate tests for this model on 1 and 2 GPUs with these changes?

@andreeahedes
Copy link
Contributor Author

andreeahedes commented Sep 4, 2023

Hi @amyeroberts , I ran test_disk_offload, test_cpu_offload and test_model_parallelism on a machine with 2 GPUs.

@SunMarc SunMarc self-requested a review September 5, 2023 13:26
Copy link
Member

@SunMarc SunMarc left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM! Just one little nit.

@HuggingFaceDocBuilderDev

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.

Copy link
Contributor

@amyeroberts amyeroberts left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for adding!

@amyeroberts amyeroberts merged commit da1af21 into huggingface:main Sep 5, 2023
@andreeahedes andreeahedes deleted the pegasus-x-no-split-modules branch September 6, 2023 16:06
parambharat pushed a commit to parambharat/transformers that referenced this pull request Sep 26, 2023
* no_split_modules

* no_split_modules

* inputs_embeds+pos same device

* update _no_split_modules

* update _no_split_modules
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.

4 participants