Skip to content

reduce memory footprint of loading pipelines from pretrained using accelerate#727

Closed
piEsposito wants to merge 0 commit intohuggingface:mainfrom
piEsposito:main
Closed

reduce memory footprint of loading pipelines from pretrained using accelerate#727
piEsposito wants to merge 0 commit intohuggingface:mainfrom
piEsposito:main

Conversation

@piEsposito
Copy link
Copy Markdown
Contributor

Should close #725.

@HuggingFaceDocBuilderDev
Copy link
Copy Markdown

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

@piEsposito piEsposito marked this pull request as ready for review October 5, 2022 12:35
@piEsposito piEsposito changed the title reduce loading pipelines from pretrained using accelerate reduce memory footprint of loading pipelines from pretrained using accelerate Oct 6, 2022
@piEsposito
Copy link
Copy Markdown
Contributor Author

@patrickvonplaten should I add tests for all pipelines or only StableDiffusion is enough?

@patrickvonplaten
Copy link
Copy Markdown
Contributor

Hey @piEsposito - super sorry I accidentally messed up your PR. I've re-opened it here: #772

I think we need to change a couple of things and also rely on changes both in Transformers and accelerate to have this working correctly.

I opened a new PR here: #772
and opened a PR on Transformers and accelerate (think both should reasonably soon release.

@patrickvonplaten
Copy link
Copy Markdown
Contributor

Very sorry about me messing up here - I made sure that you're an author of #772

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.

add device map and accelerate to DiffusionPipeline abstraction to reduce memory footprint when loading model

3 participants