feat: Make DistributedConfig Extensible (aws#5039)

* feat: Make DistributedConfig Extensible

* pylint

* Include none types when creating config jsons for safer reference

* fix: update test to account for changes

* format

* Add integ test

* pylint

* prepare release v2.240.0

* update development version to v2.240.1.dev0

* Fix key error in _send_metrics() (aws#5068)

Co-authored-by: pintaoz <[email protected]>

* fix: Added check for the presence of model package group before creating one (aws#5063)

Co-authored-by: Keshav Chandak <[email protected]>

* Use sagemaker session's s3_resource in download_folder (aws#5064)

Co-authored-by: pintaoz <[email protected]>

* remove union

* fix merge artifact

* Change dir path to distributed_drivers

* update paths

---------

Co-authored-by: ci <ci>
Co-authored-by: pintaoz-aws <[email protected]>
Co-authored-by: pintaoz <[email protected]>
Co-authored-by: Keshav Chandak <[email protected]>
Co-authored-by: Keshav Chandak <[email protected]>

Author: 5 people

src/sagemaker/modules/distributed.py

@@ -13,9 +13,12 @@
 """Distributed module."""
 from __future__ import absolute_import
 
+import os
+
+from abc import ABC, abstractmethod
 from typing import Optional, Dict, Any, List
-from pydantic import PrivateAttr
 from sagemaker.modules.utils import safe_serialize
+from sagemaker.modules.constants import SM_DRIVERS_LOCAL_PATH
 from sagemaker.modules.configs import BaseConfig
 
 
@@ -73,16 +76,37 @@ def _to_mp_hyperparameters(self) -> Dict[str, Any]:
         return hyperparameters
 
 
-class DistributedConfig(BaseConfig):
-    """Base class for distributed training configurations."""
+class DistributedConfig(BaseConfig, ABC):
+    """Abstract base class for distributed training configurations.
+
+    This class defines the interface that all distributed training configurations
+    must implement. It provides a standardized way to specify driver scripts and
+    their locations for distributed training jobs.
+    """
+
+    @property
+    @abstractmethod
+    def driver_dir(self) -> str:
+        """Directory containing the driver script.
+
+        This property should return the path to the directory containing
+        the driver script, relative to the container's working directory.
 
-    _type: str = PrivateAttr()
+        Returns:
+            str: Path to directory containing the driver script
+        """
 
-    def model_dump(self, *args, **kwargs):
-        """Dump the model to a dictionary."""
-        result = super().model_dump(*args, **kwargs)
-        result["_type"] = self._type
-        return result
+    @property
+    @abstractmethod
+    def driver_script(self) -> str:
+        """Name of the driver script.
+
+        This property should return the name of the Python script that implements
+        the distributed training driver logic.
+
+        Returns:
+            str: Name of the driver script file
+        """
 
 
 class Torchrun(DistributedConfig):
@@ -99,11 +123,27 @@ class Torchrun(DistributedConfig):
             The SageMaker Model Parallelism v2 parameters.
     """
 
-    _type: str = PrivateAttr(default="torchrun")
-
     process_count_per_node: Optional[int] = None
     smp: Optional["SMP"] = None
 
+    @property
+    def driver_dir(self) -> str:
+        """Directory containing the driver script.
+
+        Returns:
+            str: Path to directory containing the driver script
+        """
+        return os.path.join(SM_DRIVERS_LOCAL_PATH, "distributed_drivers")
+
+    @property
+    def driver_script(self) -> str:
+        """Name of the driver script.
+
+        Returns:
+            str: Name of the driver script file
+        """
+        return "torchrun_driver.py"
+
 
 class MPI(DistributedConfig):
     """MPI.
@@ -119,7 +159,23 @@ class MPI(DistributedConfig):
             The custom MPI options to use for the training job.
     """
 
-    _type: str = PrivateAttr(default="mpi")
-
     process_count_per_node: Optional[int] = None
     mpi_additional_options: Optional[List[str]] = None
+
+    @property
+    def driver_dir(self) -> str:
+        """Directory containing the driver script.
+
+        Returns:
+            str: Path to directory containing the driver script
+        """
+        return os.path.join(SM_DRIVERS_LOCAL_PATH, "distributed_drivers")
+
+    @property
+    def driver_script(self) -> str:
+        """Name of the driver script.
+
+        Returns:
+            str: Name of the driver script
+        """
+        return "mpi_driver.py"

src/sagemaker/modules/templates.py

@@ -21,17 +21,12 @@
 
 EXECUTE_BASIC_SCRIPT_DRIVER = """
 echo "Running Basic Script driver"
-$SM_PYTHON_CMD /opt/ml/input/data/sm_drivers/basic_script_driver.py
+$SM_PYTHON_CMD /opt/ml/input/data/sm_drivers/distributed_drivers/basic_script_driver.py
 """
 
-EXEUCTE_TORCHRUN_DRIVER = """
-echo "Running Torchrun driver"
-$SM_PYTHON_CMD /opt/ml/input/data/sm_drivers/torchrun_driver.py
-"""
-
-EXECUTE_MPI_DRIVER = """
-echo "Running MPI driver"
-$SM_PYTHON_CMD /opt/ml/input/data/sm_drivers/mpi_driver.py
+EXEUCTE_DISTRIBUTED_DRIVER = """
+echo "Running {driver_name} Driver"
+$SM_PYTHON_CMD /opt/ml/input/data/sm_drivers/distributed_drivers/{driver_script}
 """
 
 TRAIN_SCRIPT_TEMPLATE = """

src/sagemaker/modules/train/container_drivers/__init__.py

@@ -10,5 +10,5 @@
 # distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF
 # ANY KIND, either express or implied. See the License for the specific
 # language governing permissions and limitations under the License.
-"""Sagemaker modules container_drivers directory."""
+"""Sagemaker modules container drivers directory."""
 from __future__ import absolute_import

src/sagemaker/modules/train/container_drivers/common/__init__.py

@@ -0,0 +1,14 @@
+# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"). You
+# may not use this file except in compliance with the License. A copy of
+# the License is located at
+#
+#     http://aws.amazon.com/apache2.0/
+#
+# or in the "license" file accompanying this file. This file is
+# distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF
+# ANY KIND, either express or implied. See the License for the specific
+# language governing permissions and limitations under the License.
+"""Sagemaker modules container drivers - common directory."""
+from __future__ import absolute_import

src/sagemaker/modules/train/container_drivers/utils.py renamed to src/sagemaker/modules/train/container_drivers/common/utils.py

@@ -99,10 +99,10 @@ def read_hyperparameters_json(hyperparameters_json: Dict[str, Any] = HYPERPARAME
     return hyperparameters_dict
 
 
-def get_process_count(distributed_dict: Dict[str, Any]) -> int:
+def get_process_count(process_count: Optional[int] = None) -> int:
     """Get the number of processes to run on each node in the training job."""
     return (
-        int(distributed_dict.get("process_count_per_node", 0))
+        process_count
         or int(os.environ.get("SM_NUM_GPUS", 0))
         or int(os.environ.get("SM_NUM_NEURONS", 0))
         or 1

src/sagemaker/modules/train/container_drivers/distributed_drivers/__init__.py

@@ -0,0 +1,14 @@
+# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"). You
+# may not use this file except in compliance with the License. A copy of
+# the License is located at
+#
+#     http://aws.amazon.com/apache2.0/
+#
+# or in the "license" file accompanying this file. This file is
+# distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF
+# ANY KIND, either express or implied. See the License for the specific
+# language governing permissions and limitations under the License.
+"""Sagemaker modules container drivers - drivers directory."""
+from __future__ import absolute_import

src/sagemaker/modules/train/container_drivers/basic_script_driver.py renamed to src/sagemaker/modules/train/container_drivers/distributed_drivers/basic_script_driver.py

@@ -13,16 +13,19 @@
 """This module is the entry point for the Basic Script Driver."""
 from __future__ import absolute_import
 
+import os
 import sys
+import json
 import shlex
 
+from pathlib import Path
 from typing import List
 
-from utils import (
+sys.path.insert(0, str(Path(__file__).parent.parent))
+
+from common.utils import (  # noqa: E402 # pylint: disable=C0413,E0611
     logger,
     get_python_executable,
-    read_source_code_json,
-    read_hyperparameters_json,
     execute_commands,
     write_failure_file,
     hyperparameters_to_cli_args,
@@ -31,11 +34,10 @@
 
 def create_commands() -> List[str]:
     """Create the commands to execute."""
-    source_code = read_source_code_json()
-    hyperparameters = read_hyperparameters_json()
+    entry_script = os.environ["SM_ENTRY_SCRIPT"]
+    hyperparameters = json.loads(os.environ["SM_HPS"])
     python_executable = get_python_executable()
 
-    entry_script = source_code["entry_script"]
     args = hyperparameters_to_cli_args(hyperparameters)
     if entry_script.endswith(".py"):
         commands = [python_executable, entry_script]

src/sagemaker/modules/train/container_drivers/mpi_driver.py renamed to src/sagemaker/modules/train/container_drivers/distributed_drivers/mpi_driver.py

@@ -16,18 +16,8 @@
 import os
 import sys
 import json
+from pathlib import Path
 
-from utils import (
-    logger,
-    read_source_code_json,
-    read_distributed_json,
-    read_hyperparameters_json,
-    hyperparameters_to_cli_args,
-    get_process_count,
-    execute_commands,
-    write_failure_file,
-    USER_CODE_PATH,
-)
 from mpi_utils import (
     start_sshd_daemon,
     bootstrap_master_node,
@@ -38,6 +28,16 @@
 )
 
 
+sys.path.insert(0, str(Path(__file__).parent.parent))
+from common.utils import (  # noqa: E402 # pylint: disable=C0413,E0611
+    logger,
+    hyperparameters_to_cli_args,
+    get_process_count,
+    execute_commands,
+    write_failure_file,
+)
+
+
 def main():
     """Main function for the MPI driver script.
 
@@ -58,9 +58,9 @@ def main():
     5. Exit
 
     """
-    source_code = read_source_code_json()
-    distribution = read_distributed_json()
-    hyperparameters = read_hyperparameters_json()
+    entry_script = os.environ["SM_ENTRY_SCRIPT"]
+    distributed_config = json.loads(os.environ["SM_DISTRIBUTED_CONFIG"])
+    hyperparameters = json.loads(os.environ["SM_HPS"])
 
     sm_current_host = os.environ["SM_CURRENT_HOST"]
     sm_hosts = json.loads(os.environ["SM_HOSTS"])
@@ -77,7 +77,8 @@ def main():
 
         host_list = json.loads(os.environ["SM_HOSTS"])
         host_count = int(os.environ["SM_HOST_COUNT"])
-        process_count = get_process_count(distribution)
+        process_count = int(distributed_config["process_count_per_node"] or 0)
+        process_count = get_process_count(process_count)
 
         if process_count > 1:
             host_list = ["{}:{}".format(host, process_count) for host in host_list]
@@ -86,8 +87,8 @@ def main():
             host_count=host_count,
             host_list=host_list,
             num_processes=process_count,
-            additional_options=distribution.get("mpi_additional_options", []),
-            entry_script_path=os.path.join(USER_CODE_PATH, source_code["entry_script"]),
+            additional_options=distributed_config["mpi_additional_options"] or [],
+            entry_script_path=entry_script,
         )
 
         args = hyperparameters_to_cli_args(hyperparameters)

src/sagemaker/modules/train/container_drivers/mpi_utils.py renamed to src/sagemaker/modules/train/container_drivers/distributed_drivers/mpi_utils.py

@@ -14,12 +14,23 @@
 from __future__ import absolute_import
 
 import os
+import sys
 import subprocess
 import time
+
+from pathlib import Path
 from typing import List
 
 import paramiko
-from utils import SM_EFA_NCCL_INSTANCES, SM_EFA_RDMA_INSTANCES, get_python_executable, logger
+
+sys.path.insert(0, str(Path(__file__).parent.parent))
+
+from common.utils import (  # noqa: E402 # pylint: disable=C0413,E0611
+    SM_EFA_NCCL_INSTANCES,
+    SM_EFA_RDMA_INSTANCES,
+    get_python_executable,
+    logger,
+)
 
 FINISHED_STATUS_FILE = "/tmp/done.algo-1"
 READY_FILE = "/tmp/ready.%s"

src/sagemaker/modules/train/container_drivers/torchrun_driver.py renamed to src/sagemaker/modules/train/container_drivers/distributed_drivers/torchrun_driver.py

@@ -15,20 +15,20 @@
 
 import os
 import sys
+import json
 
+from pathlib import Path
 from typing import List, Tuple
 
-from utils import (
+sys.path.insert(0, str(Path(__file__).parent.parent))
+
+from common.utils import (  # noqa: E402 # pylint: disable=C0413,E0611
     logger,
-    read_source_code_json,
-    read_distributed_json,
-    read_hyperparameters_json,
     hyperparameters_to_cli_args,
     get_process_count,
     get_python_executable,
     execute_commands,
     write_failure_file,
-    USER_CODE_PATH,
     SM_EFA_NCCL_INSTANCES,
     SM_EFA_RDMA_INSTANCES,
 )
@@ -65,11 +65,12 @@ def setup_env():
 
 def create_commands():
     """Create the Torch Distributed command to execute"""
-    source_code = read_source_code_json()
-    distribution = read_distributed_json()
-    hyperparameters = read_hyperparameters_json()
+    entry_script = os.environ["SM_ENTRY_SCRIPT"]
+    distributed_config = json.loads(os.environ["SM_DISTRIBUTED_CONFIG"])
+    hyperparameters = json.loads(os.environ["SM_HPS"])
 
-    process_count = get_process_count(distribution)
+    process_count = int(distributed_config["process_count_per_node"] or 0)
+    process_count = get_process_count(process_count)
     host_count = int(os.environ["SM_HOST_COUNT"])
 
     torch_cmd = []
@@ -94,7 +95,7 @@ def create_commands():
             ]
         )
 
-    torch_cmd.extend([os.path.join(USER_CODE_PATH, source_code["entry_script"])])
+    torch_cmd.extend([entry_script])
 
     args = hyperparameters_to_cli_args(hyperparameters)
     torch_cmd += args

src/sagemaker/modules/train/container_drivers/scripts/__init__.py

@@ -10,5 +10,5 @@
 # distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF
 # ANY KIND, either express or implied. See the License for the specific
 # language governing permissions and limitations under the License.
-"""Sagemaker modules scripts directory."""
+"""Sagemaker modules container drivers - scripts directory."""
 from __future__ import absolute_import