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

Add cuda set_device for local run #931

Merged
merged 4 commits into from
Dec 5, 2024
Merged

Add cuda set_device for local run #931

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
b8c3905
add set local device
Dec 4, 2024
0e52700
Merge branch 'main' into rgao_fix_local_n_gpurun
rayg1234 Dec 4, 2024
f99bccd
fix test
Dec 4, 2024
20b6c1a
Merge branch 'rgao_fix_local_n_gpurun' of github.com:FAIR-Chem/fairch…
Dec 4, 2024
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
8 changes: 7 additions & 1 deletion src/fairchem/core/common/distutils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -80,7 +80,8 @@ def setup(config) -> None:
assign_device_for_local_rank(config["cpu"], config["local_rank"])
else:
# in the old code, all ranks can see all devices but need to be assigned a device equal to their local rank
# this is dangerous and should be deprecated
# this is dangerous and should be deprecated, however, FSDP still requires backwards compatibility with
# initializing this way for now so we need to keep it
torch.cuda.set_device(config["local_rank"])

dist.init_process_group(
Expand Down Expand Up @@ -123,6 +124,11 @@ def setup(config) -> None:
config["local_rank"] = int(os.environ.get("LOCAL_RANK"))
if config.get("use_cuda_visibile_devices"):
assign_device_for_local_rank(config["cpu"], config["local_rank"])
elif torch.cuda.is_available():
# in the old code, all ranks can see all devices but need to be assigned a device equal to their local rank
# this is dangerous and should be deprecated, however, FSDP still requires backwards compatibility with
# initializing this way for now so we need to keep it
torch.cuda.set_device(config["local_rank"])
dist.init_process_group(
backend=config["distributed_backend"],
rank=int(os.environ.get("RANK")),
Expand Down
Loading