Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Make fsdp ram efficient loading optional #2037

Merged
merged 6 commits into from
Oct 12, 2023
Merged

Make fsdp ram efficient loading optional #2037

merged 6 commits into from
Oct 12, 2023

Conversation

pacman100
Copy link
Contributor

What does this PR do?

  1. Make fsdp ram efficient loading optional. Certain models are having issues when handling meta devices during pre-trained model loading. Fixes Error downloading wav2vec2 model when using accelerate launch with FSDP #1948 and NotImplementedError: Cannot copy out of meta tensor; no data! - while using FSDP [works with DDP and DeepSpeed] #2031 by making the ram efficient loading optional.

@pacman100 pacman100 mentioned this pull request Oct 6, 2023
Merged
@HuggingFaceDocBuilderDev
Copy link

HuggingFaceDocBuilderDev commented Oct 6, 2023 •
edited
Loading

The documentation is not available anymore as the PR was closed or merged.

@pacman100 pacman100 requested a review from muellerzr October 6, 2023 11:14
@pacman100 pacman100 marked this pull request as ready for review October 6, 2023 11:14
muellerzr
muellerzr reviewed Oct 6, 2023
View reviewed changes
Copy link
Collaborator

@muellerzr muellerzr 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 the fix! I added a few suggestions to try and make a few points more clear, let's see if we can improve it some :)

src/accelerate/commands/config/cluster.py Outdated Show resolved Hide resolved
src/accelerate/commands/launch.py Show resolved Hide resolved
muellerzr
muellerzr approved these changes Oct 11, 2023
View reviewed changes
Copy link
Collaborator

@muellerzr muellerzr left a comment

Choose a reason for hiding this comment

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

Thanks! cc @BenjaminBossan for a secondary :)

BenjaminBossan
BenjaminBossan approved these changes Oct 12, 2023
View reviewed changes
Copy link
Member

@BenjaminBossan BenjaminBossan left a comment

Choose a reason for hiding this comment

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

Looks good, thanks Sourab. I only have two small nits, up to you if you want to address them.

src/accelerate/utils/launch.py Outdated Show resolved Hide resolved
docs/source/usage_guides/fsdp.md Outdated Show resolved Hide resolved
BenjaminBossan
BenjaminBossan approved these changes Oct 12, 2023
View reviewed changes
Copy link
Member

@BenjaminBossan BenjaminBossan left a comment

Choose a reason for hiding this comment

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

Great, thanks for enhancing the doc of the parameter.

@pacman100 pacman100 merged commit e6d96e5 into huggingface:main Oct 12, 2023
24 checks passed
@hiroshi-matsuda-rit hiroshi-matsuda-rit mentioned this pull request Oct 25, 2023
Closed
@pacman100 pacman100 mentioned this pull request Nov 8, 2023
Closed
4 tasks
@pacman100 pacman100 deleted the smangrul/make-fsdp-ram-efficient-loading-optional branch November 16, 2023 11:53
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@muellerzr muellerzr muellerzr approved these changes

@BenjaminBossan BenjaminBossan BenjaminBossan approved these changes

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

Error downloading wav2vec2 model when using accelerate launch with FSDP
4 participants
@pacman100 @HuggingFaceDocBuilderDev @BenjaminBossan @muellerzr