mindspore 2.4.10__cp310-cp310-win_amd64.whl → 2.6.0__cp310-cp310-win_amd64.whl

mindspore/communication/management.py

@@ -21,12 +21,12 @@ 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, _get_group_ranks
+    _check_mpi_envs, _set_elegant_exit_handle, _get_group_ranks, _get_comm_name_helper
 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_local_rank_size", "get_world_rank_from_group_rank", "get_comm_name",
            "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"]
 
@@ -100,10 +100,23 @@ def _set_envs():
     This takes compatibility into account because user scripts may get 'DEVICE_ID' or 'RANK_ID' envs.
     """
     os.environ["RANK_ID"] = str(get_rank())
+    os.environ["DEVICE_ID"] = str(context.get_context("device_id"))
     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 _check_hccl():
+    """Check hcll is installed needed."""
+    if not GlobalComm.CHECK_ENVS:
+        return
+    try:
+        from hccl import sys_version as hccl_version
+        v = '.'.join(hccl_version.__sys_version__.split('.')[0:2])
+        logger.debug(f"\"hccl\" wheel package version {v} is installed")
+    except Exception as e:
+        logger.error(f"Check hccl failed: {e}")
+        raise RuntimeError("\"hccl\" wheel was not installed correctly. For details, refer to the installation "
+                           "guidelines: https://www.mindspore.cn/install")
 
 
 def init(backend_name=None):
@@ -141,7 +154,7 @@ def init(backend_name=None):
             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>`_
+            <https://www.mindspore.cn/tutorials/en/master/parallel/msrun_launcher.html>`_
             for more details.
 
         >>> from mindspore.communication import init
@@ -186,6 +199,7 @@ def init(backend_name=None):
         if not host_init:
             _check_parallel_envs()
         GlobalComm.BACKEND = Backend("hccl")
+        _check_hccl()
         init_hccl()
         GlobalComm.WORLD_COMM_GROUP = HCCL_WORLD_COMM_GROUP
     elif backend_name == "nccl":
@@ -212,7 +226,7 @@ def release():
     Release distributed resource. e.g. HCCL/NCCL/MCCL.
 
     Note:
-        This method should be used after init().
+        This method should be used after init(). If not, resource will be released when program ends.
 
     Raises:
         RuntimeError: If failed to release distributed resource.
@@ -227,7 +241,7 @@ def release():
             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>`_
+            <https://www.mindspore.cn/tutorials/en/master/parallel/msrun_launcher.html>`_
             for more details.
 
         >>> from mindspore.communication import init, release
@@ -266,7 +280,7 @@ def get_rank(group=GlobalComm.WORLD_COMM_GROUP):
             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>`_
+            <https://www.mindspore.cn/tutorials/en/master/parallel/msrun_launcher.html>`_
             for more details.
 
         >>> from mindspore.communication import init, get_rank
@@ -311,13 +325,12 @@ def get_local_rank(group=GlobalComm.WORLD_COMM_GROUP):
             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>`_
+            <https://www.mindspore.cn/tutorials/en/master/parallel/msrun_launcher.html>`_
             for more details.
 
         >>> import mindspore as ms
         >>> from mindspore.communication import init, get_rank, get_local_rank
-        >>> ms.set_context(device_target="Ascend")
-        >>> ms.set_auto_parallel_context(device_num=16) # 2 server, each server with 8 NPU.
+        >>> ms.set_device(device_target="Ascend")
         >>> init()
         >>> world_rank = get_rank()
         >>> local_rank = get_local_rank()
@@ -359,12 +372,11 @@ def get_group_size(group=GlobalComm.WORLD_COMM_GROUP):
             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>`_
+            <https://www.mindspore.cn/tutorials/en/master/parallel/msrun_launcher.html>`_
             for more details.
 
         >>> import mindspore as ms
         >>> from mindspore.communication import init, get_group_size
-        >>> ms.set_auto_parallel_context(device_num=8)
         >>> init()
         >>> group_size = get_group_size()
         >>> print("group_size is: ", group_size)
@@ -406,13 +418,12 @@ def get_local_rank_size(group=GlobalComm.WORLD_COMM_GROUP):
             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>`_
+            <https://www.mindspore.cn/tutorials/en/master/parallel/msrun_launcher.html>`_
             for more details.
 
         >>> import mindspore as ms
         >>> from mindspore.communication import init, get_local_rank_size
-        >>> ms.set_context(device_target="Ascend")
-        >>> ms.set_auto_parallel_context(device_num=16) # 2 server, each server with 8 NPU.
+        >>> ms.set_device(device_target="Ascend")
         >>> init()
         >>> local_rank_size = get_local_rank_size()
         >>> print("local_rank_size is: ", local_rank_size)
@@ -456,13 +467,14 @@ def get_world_rank_from_group_rank(group, group_rank_id):
             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>`_
+            <https://www.mindspore.cn/tutorials/en/master/parallel/msrun_launcher.html>`_
             for more details.
 
         >>> import mindspore as ms
         >>> from mindspore import set_context
         >>> from mindspore.communication import init, create_group, get_world_rank_from_group_rank, get_rank
-        >>> set_context(mode=ms.GRAPH_MODE, device_target="Ascend")
+        >>> set_context(mode=ms.GRAPH_MODE)
+        >>> ms.set_device(device_target="Ascend")
         >>> init()
         >>> group = "0-4"
         >>> rank_ids = [0,4]
@@ -510,13 +522,14 @@ def get_group_rank_from_world_rank(world_rank_id, group):
             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>`_
+            <https://www.mindspore.cn/tutorials/en/master/parallel/msrun_launcher.html>`_
             for more details.
 
         >>> import mindspore as ms
         >>> from mindspore import set_context
         >>> from mindspore.communication import init, create_group, get_group_rank_from_world_rank, get_rank
-        >>> set_context(mode=ms.GRAPH_MODE, device_target="Ascend")
+        >>> set_context(mode=ms.GRAPH_MODE)
+        >>> ms.set_device(device_target="Ascend")
         >>> init()
         >>> group = "0-4"
         >>> rank_ids = [0,4]
@@ -561,14 +574,15 @@ def create_group(group, rank_ids):
             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>`_
+            <https://www.mindspore.cn/tutorials/en/master/parallel/msrun_launcher.html>`_
             for more details.
 
         >>> import mindspore as ms
         >>> from mindspore import set_context
         >>> from mindspore import ops
         >>> from mindspore.communication import init, create_group, get_rank
-        >>> set_context(mode=ms.GRAPH_MODE, device_target="Ascend")
+        >>> set_context(mode=ms.GRAPH_MODE)
+        >>> ms.set_device(device_target="Ascend")
         >>> init()
         >>> group = "0-7"
         >>> rank_ids = [0,7]
@@ -609,14 +623,15 @@ def destroy_group(group):
             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>`_
+            <https://www.mindspore.cn/tutorials/en/master/parallel/msrun_launcher.html>`_
             for more details.
 
         >>> import mindspore as ms
         >>> from mindspore import set_context
         >>> from mindspore import ops
         >>> from mindspore.communication import init, create_group, destroy_group, get_rank
-        >>> set_context(mode=ms.GRAPH_MODE, device_target="Ascend")
+        >>> set_context(mode=ms.GRAPH_MODE)
+        >>> ms.set_device(device_target="Ascend")
         >>> init()
         >>> group = "0-2"
         >>> rank_ids = [0,2]
@@ -630,6 +645,60 @@ def destroy_group(group):
     _destroy_group_helper(group)
 
 
+def get_comm_name(group=GlobalComm.WORLD_COMM_GROUP):
+    """
+    Get the communicator name of the specified collective communication group.
+
+    Note:
+        This method isn't supported in GPU and CPU versions of MindSpore.
+        This method should be used after init().
+
+    Args:
+        group (str): The communication group to work on. Normally, the group should be created by create_group,
+                     otherwise, using the default group. Default: ``GlobalComm.WORLD_COMM_GROUP`` .
+
+    Returns:
+        string, the inner communicator name of the group.
+
+    Raises:
+        TypeError: If group is not a string.
+        ValueError: If backend is invalid.
+        RuntimeError: If HCCL is not available or MindSpore is GPU/CPU version.
+
+    Supported Platforms:
+        ``Ascend``
+
+    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/tutorials/en/master/parallel/msrun_launcher.html>`_
+            for more details.
+
+        >>> import mindspore as ms
+        >>> from mindspore.communication import init, create_group, get_rank, get_comm_name
+        >>> ms.set_device(device_target="Ascend")
+        >>> init()
+        >>> world_group_comm_name = get_comm_name()
+        >>> group = "0-7"
+        >>> rank_ids = [0,7]
+        >>> if get_rank() in rank_ids:
+        ...     create_group(group, rank_ids)
+        ...     customizd_group_comm_name = get_comm_name(group)
+        ...     print("comm_name of customizd group is ", customizd_group_comm_name)
+        >>> print("comm_name of world group is: ", world_group_comm_name)
+        comm_name of customizd group is: 11.22.33.44%eth0_60000_0_0123456789101112
+        comm_name of world group is: 11.22.33.44%eth0_60000_0_1211109876543210
+    """
+    if not isinstance(group, str):
+        raise TypeError("For 'get_comm_name', the argument 'group' must be type of string, "
+                        "but got 'group' type : {}.".format(type(group)))
+    return _get_comm_name_helper(group=_get_group(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.
@@ -656,7 +725,7 @@ def get_process_group_ranks(group=GlobalComm.WORLD_COMM_GROUP):
             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>`_
+            <https://www.mindspore.cn/tutorials/en/master/parallel/msrun_launcher.html>`_
             for more details.
 
             This example should be run with 4 devices.
@@ -670,4 +739,4 @@ def get_process_group_ranks(group=GlobalComm.WORLD_COMM_GROUP):
         [0, 1, 2, 3]
 
     """
-    return _get_group_ranks(group)
+    return _get_group_ranks(group=_get_group(group))