Skip to content

Decouple device_map='auto' and tp_plan='auto' #38942

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
merged 4 commits into from
Jul 3, 2025
Merged

Decouple device_map='auto' and tp_plan='auto' #38942

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
131587c
dissociate
SunMarc Jun 20, 2025
9302204
better place
SunMarc Jun 20, 2025
72c341f
fix
SunMarc Jun 20, 2025
3e2d38e
Merge branch 'main' into device-map-tp-plan
SunMarc Jun 23, 2025
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
10 changes: 6 additions & 4 deletions src/transformers/modeling_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4307,10 +4307,12 @@ def from_pretrained(
"`tp_plan` and `device_map` are mutually exclusive. Choose either one for parallelization."
)

# If torchrun was used, make sure to TP by default. This way people don't need to change tp or device map
if device_map == "auto" and tp_plan is None and int(os.environ.get("WORLD_SIZE", 0)):
tp_plan = "auto" # device_map = "auto" in torchrun equivalent to TP plan = AUTO!
device_map = None
if device_map == "auto" and int(os.environ.get("WORLD_SIZE", 0)):
logger.info(
"You've set device_map=`auto` while triggering a distributed run with torchrun. This might lead to unexpected behavior. "
"If your plan is to load the model on each device, you should set device_map={"
": PartialState().process_index} where PartialState comes from accelerate library"
)

# We need to correctly dispatch the model on the current process device. The easiest way for this is to use a simple
# `device_map` pointing to the correct device
Expand Down