-
Notifications
You must be signed in to change notification settings - Fork 970
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.
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
[docs] Quicktour #2456
[docs] Quicktour #2456
Conversation
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. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
First pass: quicktour only
@@ -0,0 +1,128 @@ | |||
<!--Copyright 2024 The HuggingFace Team. All rights reserved. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I feel like this whole doc is just a repeat of https://moon-ci-docs.huggingface.co/docs/accelerate/pr_2456/en/concept_guides/deferring_execution no? I don't think it makes sense to have it here in the base tutorials.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think the "Deferring execution" guide shows examples of when to use main_process_first
or wait_for_everyone
such as downloading a dataset or saving/loading a state dict. This doc has a broader scope and introduces the different execution scenarios you should consider (ie, do you want to execute on one process, on a specific process).
I think introducing these types of scenarios in the tutorials gives users a nice basic understanding of how to apply it to the more specific examples, such as the ones showcased in the Deferring execution doc. Let me know what you think! 🙂
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for the rework!
This PR updates the Quicktour to focus more on the library's core offerings:
➕ organizes the Quicktour around the three main features: unified launcher,
Accelerator
class, and Big Model Inference➕ new tutorials for Execution process and TPU training
➕ adds save Transformer models from Accelerator API page to same section with saving/loading models in Add Accelerate to your code tutorial
➕ adds
on_local_main_process
,on_main_process
,on_process
, andon_local_process
from Accelerator API page to the Execution process docs➖ removes the "Common modifications of the base case" section because all these different scenarios can be overwhelming for a new user and because it is not necessarily needed to help users start quickly. These sections will be integrated as a part of the tutorials, and I'll add links to redirect users there from the Quicktour. Only exception is the Launching distributed training from a notebook because there is already a tutorial for that.