Skip to content

[Trainer] Force is_model_parallel when model is loaded in multiple GPUs using accelerate #22532

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

Merged
younesbelkada merged 6 commits into from
Apr 3, 2023
Merged

[Trainer] Force is_model_parallel when model is loaded in multiple GPUs using accelerate #22532

Changes from 5 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
1ccad27
add `is_model_parallel` arg on Trainer
younesbelkada Apr 3, 2023
316a1ec
add warning
younesbelkada Apr 3, 2023
8a0ddb9
adapt from suggestions
younesbelkada Apr 3, 2023
e394c09
revert t5 changes
younesbelkada Apr 3, 2023
d98c4c1
remove commas
younesbelkada Apr 3, 2023
5eb72b4
adapt from suggestions
younesbelkada Apr 3, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 13 additions & 0 deletions src/transformers/trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -370,6 +370,19 @@ def __init__(
else:
self.is_model_parallel = False

if (
getattr(model, "hf_device_map", None) is not None
and len(set(model.hf_device_map.values())) > 1

@sgugger sgugger Apr 3, 2023

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

Maybe actually check the number of GPUs, cause this could be one GPU and CPU here.

@younesbelkada younesbelkada Apr 3, 2023

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

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

Makes sense! Fixed in 5eb72b4

@pacman100 pacman100 Apr 3, 2023

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

Multi-device placement should only be on GPUs for naive pipelining to work, right? Offloading to CPU/disk won't work, isn't it the case?

@younesbelkada younesbelkada Apr 3, 2023

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

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

I think offloading to CPU/disk won't work yes, I am also unsure if CPU/disk offload training works out of the box with accelerate (without DeepSpeed)

and not self.is_model_parallel
):
self.is_model_parallel = True

# warn users
logger.warning(
Comment thread
younesbelkada marked this conversation as resolved.
Outdated
"You have loaded a model on multiple GPUs. `is_model_parallel` attribute will be force-set"
" to `True` to avoid any unexpected behavior such as device placement mismatching."
)

# At this stage the model is already loaded
if getattr(model, "is_loaded_in_8bit", False):
if getattr(model, "_is_int8_training_enabled", False):
Expand Down