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

Jump to bottom

Fix sagemaker DP/MP #23681

Merged
merged 4 commits into from
May 24, 2023
+9 −3
Merged

Fix sagemaker DP/MP #23681

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
7f5154a
Check for use_sagemaker_dp
muellerzr May 22, 2023
d497f8d
Add a check for is_sagemaker_mp when setting _n_gpu again. Should be …
muellerzr May 22, 2023
085e68a
Try explicit check?
muellerzr May 24, 2023
b4a2a59
Quality
muellerzr May 24, 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
4 changes: 3 additions & 1 deletion src/transformers/trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3347,7 +3347,9 @@ def _nested_gather(self, tensors, name=None):
tensors = nested_xla_mesh_reduce(tensors, name)
elif is_sagemaker_mp_enabled():
tensors = smp_gather(tensors)
elif self.args.parallel_mode == ParallelMode.DISTRIBUTED:
elif (self.args.distributed_state is not None and self.args.distributed_state.distributed_type != "NO") or (
self.args.distributed_state is None and self.local_rank != -1
):
tensors = distributed_concat(tensors)
return tensors

Expand Down
8 changes: 6 additions & 2 deletions src/transformers/training_args.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1622,6 +1622,9 @@ def _setup_devices(self) -> "torch.device":
device = torch.device("cuda", local_rank)
self._n_gpu = 1
torch.cuda.set_device(device)
elif is_sagemaker_dp_enabled():
self.distributed_state = PartialState(_use_sagemaker_dp=True)
self._n_gpu = 1
elif self.deepspeed:
# Need to do similar for Accelerator init
os.environ["ACCELERATE_USE_DEEPSPEED"] = "true"
Expand All @@ -1646,8 +1649,9 @@ def _setup_devices(self) -> "torch.device":
if is_torch_tpu_available():
device = self.distributed_state.device
self._n_gpu = 0
elif is_sagemaker_dp_enabled():
self._n_gpu = 1
elif is_sagemaker_dp_enabled() or is_sagemaker_mp_enabled():
# Already set _n_gpu
pass
elif self.distributed_state.distributed_type == DistributedType.NO:
if self.use_mps_device:
if not torch.backends.mps.is_available():
Expand Down