rm multinode eager guard tests (#48766)

* rm multinode eager guard tests

* remove unwanted tests

* reset process_mpi test

python/paddle/fluid/tests/unittests/collective/multinode/mn_dygraph_group_sharded_stage3.py

@@ -34,7 +34,6 @@ from paddle.distributed.fleet.meta_parallel.sharding.group_sharded_stage3 import
 from paddle.distributed.fleet.meta_parallel.sharding.group_sharded_utils import (
     GroupShardedScaler,
 )
-from paddle.fluid.framework import _test_eager_guard
 from paddle.nn import Linear
 
 epoch = 10
@@ -331,5 +330,4 @@ def test_stage2_stage3():
 
 
 if __name__ == '__main__':
-    with _test_eager_guard():
     test_stage2_stage3()

python/paddle/fluid/tests/unittests/collective/multinode/mn_dygraph_sharding_stage2.py

@@ -29,7 +29,6 @@ from paddle.distributed.fleet.meta_optimizers.dygraph_optimizer.sharding_optimiz
 from paddle.distributed.fleet.meta_parallel.sharding.sharding_stage2 import (
     ShardingStage2,
 )
-from paddle.fluid.framework import _test_eager_guard
 from paddle.nn import Linear
 
 seed = 2022
@@ -248,7 +247,5 @@ def test_dp_stage2():
 
 
 if __name__ == '__main__':
-    with _test_eager_guard():
-        pass
     fleet.init(is_collective=True, strategy=strategy)
     test_dp_stage2()

python/paddle/fluid/tests/unittests/collective/process_group_gloo.py

@@ -20,7 +20,6 @@ import numpy as np
 import paddle
 from paddle.fluid import core
 from paddle.fluid.dygraph.parallel import ParallelEnv
-from paddle.fluid.framework import _test_eager_guard
 
 
 class TestProcessGroupFp32(unittest.TestCase):
@@ -35,7 +34,6 @@ class TestProcessGroupFp32(unittest.TestCase):
         self.shape = (2, 10, 5)
 
     def test_create_process_group_gloo(self):
-        with _test_eager_guard():
         nranks = ParallelEnv().nranks
         rank = ParallelEnv().local_rank
         is_master = True if rank == 0 else False
@@ -175,9 +173,7 @@ class TestProcessGroupFp32(unittest.TestCase):
             task = pg.scatter(tensor_x, tensor_y, 0)
             task.wait()
         out1 = paddle.slice(tensor_x, [0], [0], [self.shape[0]])
-            out2 = paddle.slice(
-                tensor_x, [0], [self.shape[0]], [self.shape[0] * 2]
-            )
+        out2 = paddle.slice(tensor_x, [0], [self.shape[0]], [self.shape[0] * 2])
         if pg.rank() == 0:
             assert np.array_equal(tensor_y, out1)
         else:

python/paddle/fluid/tests/unittests/collective/process_group_mpi.py

@@ -28,7 +28,7 @@ from paddle.distributed.collective import (
     _set_group_map_by_name,
 )
 from paddle.fluid import core
-from paddle.fluid.framework import _set_expected_place, _test_eager_guard
+from paddle.fluid.framework import _set_expected_place
 
 ctypes.CDLL("libmpi.so", mode=ctypes.RTLD_GLOBAL)
 
@@ -444,10 +444,8 @@ class TestProcessGroup(unittest.TestCase):
         self.shape = (2, 10, 5)
 
     def test_create_process_group_mpi(self):
-        with _test_eager_guard():
         group = init_process_group()
         pg = group.process_group
-
         # test allreduce sum
         test_allreduce_sum(pg, self.shape, self.dtype)
 

python/paddle/fluid/tests/unittests/collective/process_group_nccl.py

@@ -20,7 +20,6 @@ import numpy as np
 import paddle
 import paddle.distributed as dist
 from paddle.fluid.dygraph.parallel import ParallelEnv
-from paddle.fluid.framework import _test_eager_guard
 
 
 def init_process_group(strategy=None):
@@ -44,7 +43,6 @@ class TestProcessGroupFp32(unittest.TestCase):
         self.shape = (2, 10, 5)
 
     def test_create_process_group_nccl(self):
-        with _test_eager_guard():
         device_id = paddle.distributed.ParallelEnv().dev_id
         paddle.set_device('gpu:%d' % device_id)
 
@@ -81,15 +79,11 @@ class TestProcessGroupFp32(unittest.TestCase):
         max_result = paddle.maximum(tensor_x, tensor_y)
 
         if pg.rank() == 0:
-                task = dist.all_reduce(
-                    tensor_x, dist.ReduceOp.MAX, sync_op=False
-                )
+            task = dist.all_reduce(tensor_x, dist.ReduceOp.MAX, sync_op=False)
             task.wait()
             assert np.array_equal(tensor_x, max_result)
         else:
-                task = dist.all_reduce(
-                    tensor_y, dist.ReduceOp.MAX, sync_op=False
-                )
+            task = dist.all_reduce(tensor_y, dist.ReduceOp.MAX, sync_op=False)
             task.wait()
             assert np.array_equal(tensor_y, max_result)
 
@@ -106,15 +100,11 @@ class TestProcessGroupFp32(unittest.TestCase):
         min_result = paddle.minimum(tensor_x, tensor_y)
 
         if pg.rank() == 0:
-                task = dist.all_reduce(
-                    tensor_x, dist.ReduceOp.MIN, sync_op=False
-                )
+            task = dist.all_reduce(tensor_x, dist.ReduceOp.MIN, sync_op=False)
             task.wait()
             assert np.array_equal(tensor_x, min_result)
         else:
-                task = dist.all_reduce(
-                    tensor_y, dist.ReduceOp.MIN, sync_op=False
-                )
+            task = dist.all_reduce(tensor_y, dist.ReduceOp.MIN, sync_op=False)
             task.wait()
             assert np.array_equal(tensor_y, min_result)
 
@@ -131,15 +121,11 @@ class TestProcessGroupFp32(unittest.TestCase):
         prod_result = np.multiply(x, y)
 
         if pg.rank() == 0:
-                task = dist.all_reduce(
-                    tensor_x, dist.ReduceOp.PROD, sync_op=False
-                )
+            task = dist.all_reduce(tensor_x, dist.ReduceOp.PROD, sync_op=False)
             task.wait()
             assert np.array_equal(tensor_x, prod_result)
         else:
-                task = dist.all_reduce(
-                    tensor_y, dist.ReduceOp.PROD, sync_op=False
-                )
+            task = dist.all_reduce(tensor_y, dist.ReduceOp.PROD, sync_op=False)
             task.wait()
             assert np.array_equal(tensor_y, prod_result)
 
@@ -263,9 +249,7 @@ class TestProcessGroupFp32(unittest.TestCase):
         raw_tensor_x_2 = paddle.slice(
             tensor_x, [0], [self.shape[0] // 2], [self.shape[0]]
         )
-            raw_tensor_y_1 = paddle.slice(
-                tensor_y, [0], [0], [self.shape[0] // 2]
-            )
+        raw_tensor_y_1 = paddle.slice(tensor_y, [0], [0], [self.shape[0] // 2])
         if pg.rank() == 0:
             task = pg.alltoall(tensor_x, tensor_out1)
             task.wait()
@@ -298,9 +282,7 @@ class TestProcessGroupFp32(unittest.TestCase):
         raw_tensor_x_2 = paddle.slice(
             tensor_x, [0], [self.shape[0] // 2], [self.shape[0]]
         )
-            raw_tensor_y_1 = paddle.slice(
-                tensor_y, [0], [0], [self.shape[0] // 2]
-            )
+        raw_tensor_y_1 = paddle.slice(tensor_y, [0], [0], [self.shape[0] // 2])
         if pg.rank() == 0:
             task = pg.alltoall(tensor_x, tensor_out1)
             task.wait()
@@ -352,15 +334,11 @@ class TestProcessGroupFp32(unittest.TestCase):
         max_result = paddle.maximum(tensor_x, tensor_y)
 
         if pg.rank() == 0:
-                task = dist.reduce(
-                    tensor_x, 0, dist.ReduceOp.MAX, sync_op=False
-                )
+            task = dist.reduce(tensor_x, 0, dist.ReduceOp.MAX, sync_op=False)
             task.wait()
             assert np.array_equal(tensor_x, max_result)
         else:
-                task = dist.reduce(
-                    tensor_y, 0, dist.ReduceOp.MAX, sync_op=False
-                )
+            task = dist.reduce(tensor_y, 0, dist.ReduceOp.MAX, sync_op=False)
             task.wait()
 
         print("test reduce max api ok")
@@ -376,15 +354,11 @@ class TestProcessGroupFp32(unittest.TestCase):
         min_result = paddle.minimum(tensor_x, tensor_y)
 
         if pg.rank() == 0:
-                task = dist.reduce(
-                    tensor_x, 0, dist.ReduceOp.MIN, sync_op=False
-                )
+            task = dist.reduce(tensor_x, 0, dist.ReduceOp.MIN, sync_op=False)
             task.wait()
             assert np.array_equal(tensor_x, min_result)
         else:
-                task = dist.reduce(
-                    tensor_y, 0, dist.ReduceOp.MIN, sync_op=False
-                )
+            task = dist.reduce(tensor_y, 0, dist.ReduceOp.MIN, sync_op=False)
             task.wait()
 
         print("test reduce min api ok")
@@ -400,15 +374,11 @@ class TestProcessGroupFp32(unittest.TestCase):
         prod_result = np.multiply(x, y)
 
         if pg.rank() == 0:
-                task = dist.reduce(
-                    tensor_x, 0, dist.ReduceOp.PROD, sync_op=False
-                )
+            task = dist.reduce(tensor_x, 0, dist.ReduceOp.PROD, sync_op=False)
             task.wait()
             assert np.array_equal(tensor_x, prod_result)
         else:
-                task = dist.reduce(
-                    tensor_y, 0, dist.ReduceOp.PROD, sync_op=False
-                )
+            task = dist.reduce(tensor_y, 0, dist.ReduceOp.PROD, sync_op=False)
             task.wait()
 
         print("test reduce prod api ok")
@@ -431,9 +401,7 @@ class TestProcessGroupFp32(unittest.TestCase):
             task.wait()
             paddle.device.cuda.synchronize()
         out1 = paddle.slice(tensor_x, [0], [0], [self.shape[0]])
-            out2 = paddle.slice(
-                tensor_x, [0], [self.shape[0]], [self.shape[0] * 2]
-            )
+        out2 = paddle.slice(tensor_x, [0], [self.shape[0]], [self.shape[0] * 2])
         if pg.rank() == 0:
             assert np.array_equal(tensor_y, out1)
         else: