mindspore 2.2.14__cp39-cp39-win_amd64.whl → 2.4.0__cp39-cp39-win_amd64.whl

mindspore/communication/management.py

@@ -21,13 +21,13 @@ from mindspore.communication._comm_helper import Backend, _get_rank_helper, _get
     _get_world_rank_from_group_rank_helper, _get_group_rank_from_world_rank_helper, \
     _create_group_helper, _destroy_group_helper, HCCL_WORLD_COMM_GROUP, NCCL_WORLD_COMM_GROUP, \
     MCCL_WORLD_COMM_GROUP, DEVICE_TO_BACKEND, _get_local_rank_helper, _get_local_size_helper, GlobalComm, \
-    _check_mpi_envs, _set_elegant_exit_handle
+    _check_mpi_envs, _set_elegant_exit_handle, _get_group_ranks
 from mindspore._c_expression import init_hccl, finalize_hccl, init_cluster, MSContext, ms_ctx_param
-
+from mindspore.hal.device import is_initialized
 
 __all__ = ["init", "release", "get_rank", "get_local_rank", "get_group_size",
            "get_local_rank_size", "get_world_rank_from_group_rank",
-           "get_group_rank_from_world_rank", "create_group", "destroy_group",
+           "get_group_rank_from_world_rank", "create_group", "destroy_group", "get_process_group_ranks",
            "HCCL_WORLD_COMM_GROUP", "NCCL_WORLD_COMM_GROUP", "MCCL_WORLD_COMM_GROUP"]
 
 DEFAULT_WORLD_COMM_GROUP = HCCL_WORLD_COMM_GROUP
@@ -94,6 +94,18 @@ def _check_parallel_envs():
                            "please export MINDSPORE_HCCL_CONFIG_PATH or RANK_TABLE_FILE.")
 
 
+def _set_envs():
+    """
+    Some environmental variables must be set after `init` is completed.
+    This takes compatibility into account because user scripts may get 'DEVICE_ID' or 'RANK_ID' envs.
+    """
+    os.environ["RANK_ID"] = str(get_rank())
+    if os.getenv("RANK_SIZE") is None:
+        os.environ["RANK_SIZE"] = str(get_group_size())
+    if os.getenv("DEVICE_ID") is None:
+        os.environ["DEVICE_ID"] = str(get_local_rank())
+
+
 def init(backend_name=None):
     """
     Initialize distributed backends required by communication services, e.g. ``"hccl"`` / ``"nccl"`` / ``"mccl"``.
@@ -103,6 +115,8 @@ def init(backend_name=None):
         - The full name of ``"hccl"`` is Huawei Collective Communication Library(HCCL).
         - The full name of ``"nccl"`` is NVIDIA Collective Communication Library(NCCL).
         - The full name of ``"mccl"`` is MindSpore Collective Communication Library(MCCL).
+        - In Ascend hardware platforms, ``init()`` should be set before the definition of any Tensor and Parameter,
+          and the instantiation and execution of any operation and net.
 
     Args:
         backend_name (str): Backend, using ``"hccl"`` / ``"nccl"`` / ``"mccl"``.
@@ -124,17 +138,12 @@ def init(backend_name=None):
         .. note::
             Before running the following examples, you need to configure the communication environment variables.
 
-            For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
-            Please see the `rank table Startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/rank_table.html>`_
+            For Ascend/GPU/CPU devices, it is recommended to use the msrun startup method
+            without any third-party or configuration file dependencies.
+            Please see the `msrun start up
+            <https://www.mindspore.cn/docs/zh-CN/master/model_train/parallel/msrun_launcher.html>`_
             for more details.
 
-            For the GPU devices, users need to prepare the host file and mpi, please see the `mpirun Startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/mpirun.html>`_ .
-
-            For the CPU device, users need to write a dynamic cluster startup script, please see the `Dynamic Cluster
-            Startup <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/dynamic_cluster.html>`_ .
-
         >>> from mindspore.communication import init
         >>> init()
     """
@@ -154,20 +163,26 @@ def init(backend_name=None):
     if not isinstance(backend_name, str):
         raise TypeError("For 'init', the argument 'backend_name' must be a string, "
                         "but got the type : {}".format(type(backend_name)))
+    if os.getenv("MS_ROLE") == "MS_SCHED":
+        backend_name = "mccl"
 
+    _set_elegant_exit_handle()
     if backend_name == "hccl":
         if _is_ps_mode():
-            # Use MindSpore cluster to build network for Parameter Server traning.
+            # Use MindSpore cluster to build network for Parameter Server training.
             init_cluster()
             if _is_role_sched() or _is_role_pserver():
                 raise RuntimeError("Parameter server and scheduler should use 'CPU' as backend instead of 'Ascend'")
             if _get_ps_context("worker_num") == 1:
                 GlobalComm.INITED = True
-                _set_elegant_exit_handle()
                 return
         if device_target != "Ascend":
             raise RuntimeError("For 'init', the argument 'backend_name' should be '{}' to init '{}', "
                                "but got 'hccl'.".format(DEVICE_TO_BACKEND[device_target], device_target))
+        if is_initialized(device_target):
+            logger.warning(f"For 'init' in Ascend backend, the backend is already initialized, please set it before "
+                           "the definition of any Tensor and Parameter, and the instantiation and execution of any "
+                           "operation and net, otherwise the 'init' may not take effect.")
         if not host_init:
             _check_parallel_envs()
         GlobalComm.BACKEND = Backend("hccl")
@@ -189,7 +204,7 @@ def init(backend_name=None):
                            "but got 'backend_name' : {}".format(backend_name))
 
     GlobalComm.INITED = True
-    _set_elegant_exit_handle()
+    _set_envs()
 
 
 def release():
@@ -209,17 +224,12 @@ def release():
         .. note::
             Before running the following examples, you need to configure the communication environment variables.
 
-            For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
-            Please see the `rank table Startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/rank_table.html>`_
+            For Ascend/GPU/CPU devices, it is recommended to use the msrun startup method
+            without any third-party or configuration file dependencies.
+            Please see the `msrun start up
+            <https://www.mindspore.cn/docs/zh-CN/master/model_train/parallel/msrun_launcher.html>`_
             for more details.
 
-            For the GPU devices, users need to prepare the host file and mpi, please see the `mpirun Startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/mpirun.html>`_ .
-
-            For the CPU device, users need to write a dynamic cluster startup script, please see the `Dynamic Cluster
-            Startup <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/dynamic_cluster.html>`_ .
-
         >>> from mindspore.communication import init, release
         >>> init()
         >>> release()
@@ -253,17 +263,12 @@ def get_rank(group=GlobalComm.WORLD_COMM_GROUP):
         .. note::
             Before running the following examples, you need to configure the communication environment variables.
 
-            For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
-            Please see the `rank table Startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/rank_table.html>`_
+            For Ascend/GPU/CPU devices, it is recommended to use the msrun startup method
+            without any third-party or configuration file dependencies.
+            Please see the `msrun start up
+            <https://www.mindspore.cn/docs/zh-CN/master/model_train/parallel/msrun_launcher.html>`_
             for more details.
 
-            For the GPU devices, users need to prepare the host file and mpi, please see the `mpirun Startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/mpirun.html>`_ .
-
-            For the CPU device, users need to write a dynamic cluster startup script, please see the `Dynamic Cluster
-            Startup <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/dynamic_cluster.html>`_ .
-
         >>> from mindspore.communication import init, get_rank
         >>> init()
         >>> rank_id = get_rank()
@@ -297,23 +302,18 @@ def get_local_rank(group=GlobalComm.WORLD_COMM_GROUP):
         RuntimeError: If HCCL is not available or MindSpore is GPU/CPU version.
 
     Supported Platforms:
-        ``Ascend``
+        ``Ascend`` ``GPU`` ``CPU``
 
     Examples:
         .. note::
             Before running the following examples, you need to configure the communication environment variables.
 
-            For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
-            Please see the `rank table Startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/rank_table.html>`_
+            For Ascend/GPU/CPU devices, it is recommended to use the msrun startup method
+            without any third-party or configuration file dependencies.
+            Please see the `msrun start up
+            <https://www.mindspore.cn/docs/zh-CN/master/model_train/parallel/msrun_launcher.html>`_
             for more details.
 
-            For the GPU devices, users need to prepare the host file and mpi, please see the `mpirun Startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/mpirun.html>`_ .
-
-            For the CPU device, users need to write a dynamic cluster startup script, please see the `Dynamic Cluster
-            Startup <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/dynamic_cluster.html>`_ .
-
         >>> import mindspore as ms
         >>> from mindspore.communication import init, get_rank, get_local_rank
         >>> ms.set_context(device_target="Ascend")
@@ -356,17 +356,12 @@ def get_group_size(group=GlobalComm.WORLD_COMM_GROUP):
         .. note::
             Before running the following examples, you need to configure the communication environment variables.
 
-            For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
-            Please see the `rank table Startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/rank_table.html>`_
+            For Ascend/GPU/CPU devices, it is recommended to use the msrun startup method
+            without any third-party or configuration file dependencies.
+            Please see the `msrun start up
+            <https://www.mindspore.cn/docs/zh-CN/master/model_train/parallel/msrun_launcher.html>`_
             for more details.
 
-            For the GPU devices, users need to prepare the host file and mpi, please see the `mpirun Startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/mpirun.html>`_ .
-
-            For the CPU device, users need to write a dynamic cluster startup script, please see the `Dynamic Cluster
-            Startup <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/dynamic_cluster.html>`_ .
-
         >>> import mindspore as ms
         >>> from mindspore.communication import init, get_group_size
         >>> ms.set_auto_parallel_context(device_num=8)
@@ -402,23 +397,18 @@ def get_local_rank_size(group=GlobalComm.WORLD_COMM_GROUP):
         RuntimeError: If HCCL is not available or MindSpore is GPU/CPU version.
 
     Supported Platforms:
-        ``Ascend``
+        ``Ascend`` ``GPU`` ``CPU``
 
     Examples:
         .. note::
             Before running the following examples, you need to configure the communication environment variables.
 
-            For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
-            Please see the `rank table Startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/rank_table.html>`_
+            For Ascend/GPU/CPU devices, it is recommended to use the msrun startup method
+            without any third-party or configuration file dependencies.
+            Please see the `msrun start up
+            <https://www.mindspore.cn/docs/zh-CN/master/model_train/parallel/msrun_launcher.html>`_
             for more details.
 
-            For the GPU devices, users need to prepare the host file and mpi, please see the `mpirun Startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/mpirun.html>`_ .
-
-            For the CPU device, users need to write a dynamic cluster startup script, please see the `Dynamic Cluster
-            Startup <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/dynamic_cluster.html>`_ .
-
         >>> import mindspore as ms
         >>> from mindspore.communication import init, get_local_rank_size
         >>> ms.set_context(device_target="Ascend")
@@ -457,23 +447,18 @@ def get_world_rank_from_group_rank(group, group_rank_id):
         RuntimeError: If HCCL is not available or MindSpore is GPU/CPU version.
 
     Supported Platforms:
-        ``Ascend``
+        ``Ascend`` ``GPU`` ``CPU``
 
     Examples:
         .. note::
             Before running the following examples, you need to configure the communication environment variables.
 
-            For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
-            Please see the `rank table Startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/rank_table.html>`_
+            For Ascend/GPU/CPU devices, it is recommended to use the msrun startup method
+            without any third-party or configuration file dependencies.
+            Please see the `msrun start up
+            <https://www.mindspore.cn/docs/zh-CN/master/model_train/parallel/msrun_launcher.html>`_
             for more details.
 
-            For the GPU devices, users need to prepare the host file and mpi, please see the `mpirun Startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/mpirun.html>`_
-
-            For the CPU device, users need to write a dynamic cluster startup script, please see the `Dynamic Cluster
-            Startup <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/dynamic_cluster.html>`_ .
-
         >>> import mindspore as ms
         >>> from mindspore import set_context
         >>> from mindspore.communication import init, create_group, get_world_rank_from_group_rank, get_rank
@@ -516,23 +501,18 @@ def get_group_rank_from_world_rank(world_rank_id, group):
         RuntimeError: If HCCL is not available or MindSpore is GPU/CPU version.
 
     Supported Platforms:
-        ``Ascend``
+        ``Ascend`` ``GPU`` ``CPU``
 
     Examples:
         .. note::
             Before running the following examples, you need to configure the communication environment variables.
 
-            For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
-            Please see the `rank table Startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/rank_table.html>`_
+            For Ascend/GPU/CPU devices, it is recommended to use the msrun startup method
+            without any third-party or configuration file dependencies.
+            Please see the `msrun start up
+            <https://www.mindspore.cn/docs/zh-CN/master/model_train/parallel/msrun_launcher.html>`_
             for more details.
 
-            For the GPU devices, users need to prepare the host file and mpi, please see the `mpirun Startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/mpirun.html>`_
-
-            For the CPU device, users need to write a dynamic cluster startup script, please see the `Dynamic Cluster
-            Startup <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/dynamic_cluster.html>`_ .
-
         >>> import mindspore as ms
         >>> from mindspore import set_context
         >>> from mindspore.communication import init, create_group, get_group_rank_from_world_rank, get_rank
@@ -572,26 +552,21 @@ def create_group(group, rank_ids):
         RuntimeError: If HCCL is not available or MindSpore is GPU/CPU version.
 
     Supported Platforms:
-        ``Ascend``
+        ``Ascend`` ``GPU`` ``CPU``
 
     Examples:
         .. note::
             Before running the following examples, you need to configure the communication environment variables.
 
-            For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
-            Please see the `rank table Startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/rank_table.html>`_
+            For Ascend/GPU/CPU devices, it is recommended to use the msrun startup method
+            without any third-party or configuration file dependencies.
+            Please see the `msrun start up
+            <https://www.mindspore.cn/docs/zh-CN/master/model_train/parallel/msrun_launcher.html>`_
             for more details.
 
-            For the GPU devices, users need to prepare the host file and mpi, please see the `mpirun Startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/mpirun.html>`_ .
-
-            For the CPU device, users need to write a dynamic cluster startup script, please see the `Dynamic Cluster
-            Startup <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/dynamic_cluster.html>`_ .
-
         >>> import mindspore as ms
         >>> from mindspore import set_context
-        >>> import mindspore.ops as ops
+        >>> from mindspore import ops
         >>> from mindspore.communication import init, create_group, get_rank
         >>> set_context(mode=ms.GRAPH_MODE, device_target="Ascend")
         >>> init()
@@ -625,26 +600,21 @@ def destroy_group(group):
         RuntimeError: If HCCL is not available or MindSpore is GPU/CPU version.
 
     Supported Platforms:
-        ``Ascend``
+        ``Ascend`` ``GPU`` ``CPU``
 
     Examples:
         .. note::
             Before running the following examples, you need to configure the communication environment variables.
 
-            For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
-            Please see the `rank table startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/rank_table.html>`_
+            For Ascend/GPU/CPU devices, it is recommended to use the msrun startup method
+            without any third-party or configuration file dependencies.
+            Please see the `msrun start up
+            <https://www.mindspore.cn/docs/zh-CN/master/model_train/parallel/msrun_launcher.html>`_
             for more details.
 
-            For the GPU devices, users need to prepare the host file and mpi, please see the `mpirun startup
-            <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/mpirun.html>`_ .
-
-            For the CPU device, users need to write a dynamic cluster startup script, please see the `Dynamic Cluster
-            Startup <https://www.mindspore.cn/tutorials/experts/en/r2.2/parallel/dynamic_cluster.html>`_ .
-
         >>> import mindspore as ms
         >>> from mindspore import set_context
-        >>> import mindspore.ops as ops
+        >>> from mindspore import ops
         >>> from mindspore.communication import init, create_group, destroy_group, get_rank
         >>> set_context(mode=ms.GRAPH_MODE, device_target="Ascend")
         >>> init()
@@ -658,3 +628,46 @@ def destroy_group(group):
         raise TypeError("For 'destroy_group', the argument 'group' must be type of string, "
                         "but got 'group' type : {}.".format(type(group)))
     _destroy_group_helper(group)
+
+
+def get_process_group_ranks(group=GlobalComm.WORLD_COMM_GROUP):
+    """
+    Gets the ranks of the specific group and returns the process ranks in the communication group as a list.
+
+    Args:
+        group (str, optional): The communication group to work on. Default: ``GlobalComm.WORLD_COMM_GROUP`` , which
+            means ``"hccl_world_group"`` in Ascend, and ``"nccl_world_group"`` in GPU.
+
+    Returns:
+        List (List[int]), List of process ranks in the specified communication group.
+
+    Raises:
+        TypeError: If the `group` is not a str.
+        RuntimeError: If device target is invalid, or backend is invalid, or distributed initialization fails.
+
+    Supported Platforms:
+        ``Ascend`` ``GPU``
+
+    Examples:
+        .. note::
+            Before running the following examples, you need to configure the communication environment variables.
+
+            For Ascend/GPU/CPU devices, it is recommended to use the msrun startup method
+            without any third-party or configuration file dependencies.
+
+            Please see the `msrun start up
+            <https://www.mindspore.cn/docs/zh-CN/master/model_train/parallel/msrun_launcher.html>`_
+            for more details.
+
+            This example should be run with 4 devices.
+
+        >>> import numpy as np
+        >>> from mindspore.communication import init, get_process_group_ranks
+        >>>
+        >>> init()
+        >>> output = get_process_group_ranks()
+        >>> print(output)
+        [0, 1, 2, 3]
+
+    """
+    return _get_group_ranks(group)