mindspore 2.0.0a0__cp38-cp38-win_amd64.whl → 2.0.0rc1__cp38-cp38-win_amd64.whl

mindspore/communication/__init__.py

@@ -19,11 +19,11 @@ Note that the APIs in the following list need to preset communication environmen
 
 For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
 Please see the `Ascend tutorial
-<https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_ascend.html#preparations>`_
+<https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_ascend.html#preparations>`_
 for more details.
 
 For the GPU devices, users need to prepare the host file and mpi, please see the `GPU tutorial
-<https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_gpu.html#preparation>`_ .
+<https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_gpu.html#preparation>`_ .
 """
 
 from mindspore.communication.management import GlobalComm, init, release, get_rank, \

mindspore/communication/_comm_helper.py

@@ -16,6 +16,10 @@
 
 import os
 import ctypes
+
+import sys
+from sys import excepthook
+
 from mindspore import context
 from mindspore.parallel._ps_context import _is_role_worker, _is_role_pserver, _is_role_sched, _is_ps_mode,\
                                            _get_ps_context
@@ -29,7 +33,7 @@ def hccl_load_lib():
     """load hccl lib"""
     try:
         base_dir = os.path.dirname(os.path.realpath(__file__))
-        lib_path = os.path.join(base_dir, "../lib/plugin", HCCL_LIB)
+        lib_path = os.path.join(base_dir, "../lib/plugin/ascend", HCCL_LIB)
         ctypes.CDLL(lib_path)
     except Exception:
         raise RuntimeError('Get hccl lib error.')
@@ -145,8 +149,6 @@ def _check_bypass_rank_id_and_size():
 
 def _set_elegant_exit_handle():
     if _is_role_worker() or _is_role_pserver() or _is_role_sched():
-        import sys
-        from sys import excepthook
         sys.excepthook = lambda *args: (set_cluster_exit_with_exception(), excepthook(*args))
 
 
@@ -350,7 +352,12 @@ def _create_group_helper(group, rank_ids):
     if _hccl_test():
         hccl.create_group(group, rank_size, rank_ids)
     else:
-        CollectiveManager.get_instance().create_group(group, rank_ids)
+        result = CollectiveManager.get_instance().create_group(group, rank_ids)
+        if not result:
+            raise RuntimeError("Failed to create communication group for {} with rank ids {}. "
+                               "If NCCL is used, 'export NCCL_DEBUG=INFO' "
+                               "is suggested before launching jobs.".format(group, rank_ids))
+
     _ExistingGroup.ITEMS[group] = rank_ids
 
 

mindspore/communication/management.py

@@ -109,7 +109,7 @@ def init(backend_name=None):
                       have not been exported when backend is HCCL.
 
     Supported Platforms:
-        ``Ascend`` ``GPU``
+        ``Ascend`` ``GPU`` ``CPU``
 
     Examples:
         .. note::
@@ -117,11 +117,11 @@ def init(backend_name=None):
 
             For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
             Please see the `Ascend tutorial
-            <https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_ascend.html#preparations>`_
+            <https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_ascend.html#preparations>`_
             for more details.
 
             For the GPU devices, users need to prepare the host file and mpi, please see the `GPU tutorial
-            <https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_gpu.html#preparation>`_ .
+            <https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_gpu.html#preparation>`_ .
 
         >>> from mindspore.communication import init
         >>> init()
@@ -151,6 +151,7 @@ def init(backend_name=None):
                 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 'Ascend' to init hccl, "
@@ -192,11 +193,11 @@ def release():
 
             For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
             Please see the `Ascend tutorial
-            <https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_ascend.html#preparations>`_
+            <https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_ascend.html#preparations>`_
             for more details.
 
             For the GPU devices, users need to prepare the host file and mpi, please see the `GPU tutorial
-            <https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_gpu.html#preparation>`_ .
+            <https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_gpu.html#preparation>`_ .
 
         >>> from mindspore.communication import init, release
         >>> init()
@@ -225,7 +226,7 @@ def get_rank(group=GlobalComm.WORLD_COMM_GROUP):
         RuntimeError: If HCCL/NCCL is not available.
 
     Supported Platforms:
-        ``Ascend`` ``GPU``
+        ``Ascend`` ``GPU`` ``CPU``
 
     Examples:
         .. note::
@@ -233,11 +234,11 @@ def get_rank(group=GlobalComm.WORLD_COMM_GROUP):
 
             For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
             Please see the `Ascend tutorial
-            <https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_ascend.html#preparations>`_
+            <https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_ascend.html#preparations>`_
             for more details.
 
             For the GPU devices, users need to prepare the host file and mpi, please see the `GPU tutorial
-            <https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_gpu.html#preparation>`_ .
+            <https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_gpu.html#preparation>`_ .
 
         >>> from mindspore.communication import init, get_rank
         >>> init()
@@ -280,11 +281,11 @@ def get_local_rank(group=GlobalComm.WORLD_COMM_GROUP):
 
             For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
             Please see the `Ascend tutorial
-            <https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_ascend.html#preparations>`_
+            <https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_ascend.html#preparations>`_
             for more details.
 
             For the GPU devices, users need to prepare the host file and mpi, please see the `GPU tutorial
-            <https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_gpu.html#preparation>`_ .
+            <https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_gpu.html#preparation>`_ .
 
         >>> import mindspore as ms
         >>> from mindspore.communication.management import init, get_rank, get_local_rank
@@ -322,7 +323,7 @@ def get_group_size(group=GlobalComm.WORLD_COMM_GROUP):
         RuntimeError: If HCCL/NCCL is not available.
 
     Supported Platforms:
-        ``Ascend`` ``GPU``
+        ``Ascend`` ``GPU`` ``CPU``
 
     Examples:
         .. note::
@@ -330,11 +331,11 @@ def get_group_size(group=GlobalComm.WORLD_COMM_GROUP):
 
             For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
             Please see the `Ascend tutorial
-            <https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_ascend.html#preparations>`_
+            <https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_ascend.html#preparations>`_
             for more details.
 
             For the GPU devices, users need to prepare the host file and mpi, please see the `GPU tutorial
-            <https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_gpu.html#preparation>`_ .
+            <https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_gpu.html#preparation>`_ .
 
         >>> import mindspore as ms
         >>> from mindspore.communication.management import init, get_group_size
@@ -379,11 +380,11 @@ def get_local_rank_size(group=GlobalComm.WORLD_COMM_GROUP):
 
             For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
             Please see the `Ascend tutorial
-            <https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_ascend.html#preparations>`_
+            <https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_ascend.html#preparations>`_
             for more details.
 
             For the GPU devices, users need to prepare the host file and mpi, please see the `GPU tutorial
-            <https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_gpu.html#preparation>`_ .
+            <https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_gpu.html#preparation>`_ .
 
         >>> import mindspore as ms
         >>> from mindspore.communication.management import init, get_local_rank_size
@@ -431,11 +432,11 @@ def get_world_rank_from_group_rank(group, group_rank_id):
 
             For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
             Please see the `Ascend tutorial
-            <https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_ascend.html#preparations>`_
+            <https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_ascend.html#preparations>`_
             for more details.
 
             For the GPU devices, users need to prepare the host file and mpi, please see the `GPU tutorial
-            <https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_gpu.html#preparation>`_ .
+            <https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_gpu.html#preparation>`_ .
 
         >>> from mindspore import set_context
         >>> from mindspore.communication.management import init, create_group, get_world_rank_from_group_rank
@@ -485,11 +486,11 @@ def get_group_rank_from_world_rank(world_rank_id, group):
 
             For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
             Please see the `Ascend tutorial
-            <https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_ascend.html#preparations>`_
+            <https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_ascend.html#preparations>`_
             for more details.
 
             For the GPU devices, users need to prepare the host file and mpi, please see the `GPU tutorial
-            <https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_gpu.html#preparation>`_ .
+            <https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_gpu.html#preparation>`_ .
 
         >>> from mindspore import set_context
         >>> from mindspore.communication.management import init, create_group, get_group_rank_from_world_rank
@@ -536,11 +537,11 @@ def create_group(group, rank_ids):
 
             For the Ascend devices, users need to prepare the rank table, set rank_id and device_id.
             Please see the `Ascend tutorial
-            <https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_ascend.html#preparations>`_
+            <https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_ascend.html#preparations>`_
             for more details.
 
             For the GPU devices, users need to prepare the host file and mpi, please see the `GPU tutorial
-            <https://www.mindspore.cn/tutorials/experts/en/r2.0.0-alpha/parallel/train_gpu.html#preparation>`_ .
+            <https://www.mindspore.cn/tutorials/experts/en/r2.0/parallel/train_gpu.html#preparation>`_ .
 
         >>> from mindspore import set_context
         >>> import mindspore.ops as ops