Skip to content

feature: Add torch_distributed support for Trainium #248

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 2 commits into from
Oct 20, 2022
Merged

feature: Add torch_distributed support for Trainium #248

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
305c6eb
feature: Add torch_distributed support for Trainium instaces
satishpasumarthi Oct 19, 2022
a491542
update sagemaker-training package requirement
satishpasumarthi Oct 20, 2022
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
2 changes: 1 addition & 1 deletion setup.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -53,7 +53,7 @@ def read(fname):
'Programming Language :: Python :: 3.9',
],

install_requires=['retrying', 'sagemaker-training>=4.2.0', 'six>=1.12.0'],
install_requires=['retrying', 'sagemaker-training>=4.3.0', 'six>=1.12.0'],
extras_require={
'test': test_dependencies
},
Expand Down
8 changes: 8 additions & 0 deletions src/sagemaker_pytorch_container/training.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,6 +24,7 @@
LAUNCH_MPI_ENV_NAME = 'sagemaker_mpi_enabled'
LAUNCH_PYTORCH_DDP_ENV_NAME = "sagemaker_pytorch_ddp_enabled"
LAUNCH_PYTORCH_XLA_ENV_NAME = "sagemaker_pytorch_xla_multi_worker_enabled"
LAUNCH_TORCH_DISTRIBUTED_ENV_NAME = "sagemaker_torch_distributed_enabled"

logger = logging.getLogger(__name__)

Expand Down Expand Up @@ -65,6 +66,10 @@ def train(training_environment):
pytorch_xla_enabled = training_environment.additional_framework_parameters.get(
LAUNCH_PYTORCH_XLA_ENV_NAME, False
)

torch_distributed_enabled = training_environment.additional_framework_parameters.get(
LAUNCH_TORCH_DISTRIBUTED_ENV_NAME, False
)
# default scenario
runner_type = runner.ProcessRunnerType

Expand All @@ -74,6 +79,9 @@ def train(training_environment):
elif pytorch_ddp_enabled:
runner_type = runner.SMDataParallelRunnerType
logger.info('Invoking SMDataParallel for native PT DDP job')
elif torch_distributed_enabled:
runner_type = runner.TorchDistributedRunnerType
logger.info('Invoking TorchDistributed...')
elif smdataparallel_enabled:
runner_type = runner.SMDataParallelRunnerType
logger.info('Invoking SMDataParallel')
Expand Down
16 changes: 16 additions & 0 deletions test/unit/test_train.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -120,6 +120,22 @@ def test_train_pytorch_ddp(run_module, training_env):
)


@patch("sagemaker_training.entry_point.run")
@patch('socket.gethostbyname', MagicMock())
def test_train_torch_distributed(run_module, training_env):
training_env.additional_framework_parameters["sagemaker_torch_distributed_enabled"] = True

train(training_env)
run_module.assert_called_with(
uri=training_env.module_dir,
user_entry_point=training_env.user_entry_point,
args=training_env.to_cmd_args(),
env_vars=training_env.to_env_vars(),
capture_error=True,
runner_type=runner.TorchDistributedRunnerType,
)


@patch("sagemaker_training.entry_point.run")
@patch('socket.gethostbyname', MagicMock())
def test_train_pytorch_xla_distributed(run_module, training_env):
Expand Down