Fix test and doc (#44735)

* fix test and doc

python/paddle/fluid/tests/unittests/test_sparse_norm_op.py

@@ -18,7 +18,6 @@ import numpy as np
 import paddle
 from paddle.incubate.sparse import nn
 import paddle.fluid as fluid
-from paddle.fluid.framework import _test_eager_guard
 import copy
 
 
@@ -26,7 +25,6 @@ class TestSparseBatchNorm(unittest.TestCase):
 
     def test(self):
         fluid.set_flags({"FLAGS_retain_grad_for_all_tensor": True})
-        with _test_eager_guard():
         paddle.seed(0)
         channels = 4
         shape = [2, 3, 6, 6, channels]
@@ -64,7 +62,6 @@ class TestSparseBatchNorm(unittest.TestCase):
         fluid.set_flags({"FLAGS_retain_grad_for_all_tensor": False})
 
     def test_error_layout(self):
-        with _test_eager_guard():
         with self.assertRaises(ValueError):
             shape = [2, 3, 6, 6, 3]
             x = paddle.randn(shape)
@@ -74,7 +71,6 @@ class TestSparseBatchNorm(unittest.TestCase):
             sparse_batch_norm(sparse_x)
 
     def test2(self):
-        with _test_eager_guard():
         paddle.seed(123)
         channels = 3
         x_data = paddle.randn((1, 6, 6, 6, channels)).astype('float32')
@@ -82,23 +78,30 @@ class TestSparseBatchNorm(unittest.TestCase):
         sparse_x = dense_x.to_sparse_coo(4)
         batch_norm = paddle.incubate.sparse.nn.BatchNorm(channels)
         batch_norm_out = batch_norm(sparse_x)
-            print(batch_norm_out.shape)
+        dense_bn = paddle.nn.BatchNorm1D(channels)
+        dense_x = dense_x.reshape((-1, dense_x.shape[-1]))
+        dense_out = dense_bn(dense_x)
+        assert np.allclose(dense_out.numpy(), batch_norm_out.values().numpy())
         # [1, 6, 6, 6, 3]
 
 
 class TestSyncBatchNorm(unittest.TestCase):
 
     def test_sync_batch_norm(self):
-        with _test_eager_guard():
         x = np.array([[[[0.3, 0.4], [0.3, 0.07]],
                        [[0.83, 0.37], [0.18, 0.93]]]]).astype('float32')
         x = paddle.to_tensor(x)
-            x = x.to_sparse_coo(len(x.shape) - 1)
+        sparse_x = x.to_sparse_coo(len(x.shape) - 1)
 
         if paddle.is_compiled_with_cuda():
-                sync_batch_norm = nn.SyncBatchNorm(2)
-                hidden1 = sync_batch_norm(x)
-                print(hidden1)
+            sparse_sync_bn = nn.SyncBatchNorm(2)
+            sparse_hidden = sparse_sync_bn(sparse_x)
+
+            dense_sync_bn = paddle.nn.SyncBatchNorm(2)
+            x = x.reshape((-1, x.shape[-1]))
+            dense_hidden = dense_sync_bn(x)
+            assert np.allclose(sparse_hidden.values().numpy(),
+                               dense_hidden.numpy())
 
     def test_convert(self):
         base_model = paddle.nn.Sequential(nn.Conv3D(3, 5, 3), nn.BatchNorm(5),

python/paddle/incubate/sparse/nn/layer/norm.py

@@ -229,6 +229,7 @@ class SyncBatchNorm(paddle.nn.SyncBatchNorm):
 
     Shapes:
         input: Tensor that the dimension from 2 to 5.
+
         output: Tensor with the same shape as input.
 
     Examples:
@@ -278,7 +279,7 @@ class SyncBatchNorm(paddle.nn.SyncBatchNorm):
 
     @classmethod
     def convert_sync_batchnorm(cls, layer):
-        """
+        r"""
         Helper function to convert :class: `paddle.incubate.sparse.nn.BatchNorm` layers in the model to :class: `paddle.incubate.sparse.nn.SyncBatchNorm` layers.
 
         Parameters:
@@ -290,13 +291,14 @@ class SyncBatchNorm(paddle.nn.SyncBatchNorm):
         Examples:
 
             .. code-block:: python
+
                 import paddle
                 import paddle.incubate.sparse.nn as nn
 
                 model = paddle.nn.Sequential(nn.Conv3D(3, 5, 3), nn.BatchNorm(5))
                 sync_model = nn.SyncBatchNorm.convert_sync_batchnorm(model)
-
         """
+
         layer_output = layer
         if isinstance(layer, _BatchNormBase):
             if layer._weight_attr != None and not isinstance(