diff --git a/paddle/fluid/operators/generator/generate_sparse_op.py b/paddle/fluid/operators/generator/generate_sparse_op.py
index 6c9f70b9283f88779ca175d89a01c87fd8296d25..10ee034ff3b47305a323850a07ec5c901c98b058 100644
--- a/paddle/fluid/operators/generator/generate_sparse_op.py
+++ b/paddle/fluid/operators/generator/generate_sparse_op.py
@@ -82,6 +82,8 @@ def main(op_yaml_path, backward_yaml_path, output_op_path, output_arg_map_path):
     backward_op_dict = to_named_dict(backward_ops)
 
     for op in ops:
+        if op['name'][-1] == '_':
+            op['name'] = op['name'][:-1]
         op['op_name'] = SPARSE_OP_PREFIX + op['name']
         op['name'] = op['op_name']
         if op["backward"] is not None:
diff --git a/paddle/phi/api/yaml/sparse_backward.yaml b/paddle/phi/api/yaml/sparse_backward.yaml
index 3e654210b9008280e12a9429c17a952991e16002..949a6c4c19b12cdaa636b41f96dc38e9db5c9e0f 100644
--- a/paddle/phi/api/yaml/sparse_backward.yaml
+++ b/paddle/phi/api/yaml/sparse_backward.yaml
@@ -101,7 +101,7 @@
            atanh_csr_grad {sparse_csr, sparse_csr -> sparse_csr}
 
 - backward_op : batch_norm_grad
-  forward : batch_norm (Tensor x, Tensor mean, Tensor variance, Tensor scale, Tensor bias, bool is_test, float momentum, float epsilon, str data_layout, bool use_global_stats, bool trainable_statistics) -> Tensor(out), Tensor(mean_out), Tensor(variance_out), Tensor(saved_mean), Tensor(saved_variance), Tensor(reserve_space)
+  forward : batch_norm_ (Tensor x, Tensor mean, Tensor variance, Tensor scale, Tensor bias, bool is_test, float momentum, float epsilon, str data_layout, bool use_global_stats, bool trainable_statistics) -> Tensor(out), Tensor(mean_out), Tensor(variance_out), Tensor(saved_mean), Tensor(saved_variance), Tensor(reserve_space)
   args : (Tensor x, Tensor scale, Tensor bias, Tensor mean_out, Tensor variance_out, Tensor saved_mean, Tensor saved_variance, Tensor reserve_space, Tensor out_grad, float momentum, float epsilon, str data_layout, bool is_test, bool use_global_stats, bool trainable_statistics)
   output : Tensor(x_grad), Tensor(scale_grad), Tensor(bias_grad)
   infer_meta :
diff --git a/paddle/phi/api/yaml/sparse_ops.yaml b/paddle/phi/api/yaml/sparse_ops.yaml
index 545042b6f073e447b4c6e7e342c0be82e791e3b6..2208d34ecf4cda1328342e9f53b5cf00fd5aed08 100644
--- a/paddle/phi/api/yaml/sparse_ops.yaml
+++ b/paddle/phi/api/yaml/sparse_ops.yaml
@@ -87,7 +87,7 @@
     layout : x
   backward : atanh_grad
 
-- op : batch_norm
+- op : batch_norm_
   args : (Tensor x, Tensor mean, Tensor variance, Tensor scale, Tensor bias, bool is_test, float momentum, float epsilon, str data_layout, bool use_global_stats, bool trainable_statistics)
   output : Tensor(out), Tensor(mean_out), Tensor(variance_out), Tensor(saved_mean), Tensor(saved_variance), Tensor(reserve_space)
   infer_meta :
@@ -95,7 +95,7 @@
   kernel :
     func : batch_norm_coo {sparse_coo, dense, dense, dense, dense -> sparse_coo, dense, dense, dense, dense, dense}
     data_type : x
-  view : (mean -> mean_out), (variance -> variance_out)
+  inplace : (mean -> mean_out), (variance -> variance_out)
   backward : batch_norm_grad
 
 - op : cast
diff --git a/paddle/phi/kernels/sparse/batch_norm_kernel.h b/paddle/phi/kernels/sparse/batch_norm_kernel.h
index 03e621cc65325bf2394c255ae97b9874020751a0..41656280c500e3522aec0656d68369711c157526 100644
--- a/paddle/phi/kernels/sparse/batch_norm_kernel.h
+++ b/paddle/phi/kernels/sparse/batch_norm_kernel.h
@@ -23,24 +23,24 @@ namespace phi {
 namespace sparse {
 
 template <typename T, typename Context>
-void BatchNormKernel(const Context& dev_ctx,
-                     const SparseCooTensor& x,
-                     const DenseTensor& scale,
-                     const DenseTensor& bias,
-                     const DenseTensor& mean,
-                     const DenseTensor& variance,
-                     float momentum,
-                     float epsilon,
-                     const std::string& data_layout,
-                     bool is_test,
-                     bool use_global_stats,
-                     bool trainable_statistics,
-                     SparseCooTensor* y,
-                     DenseTensor* mean_out,
-                     DenseTensor* variance_out,
-                     DenseTensor* saved_mean,
-                     DenseTensor* saved_variance,
-                     DenseTensor* reserve_space);
+void BatchNormCooKernel(const Context& dev_ctx,
+                        const SparseCooTensor& x,
+                        const DenseTensor& mean,
+                        const DenseTensor& variance,
+                        const DenseTensor& scale,
+                        const DenseTensor& bias,
+                        bool is_test,
+                        float momentum,
+                        float epsilon,
+                        const std::string& data_layout,
+                        bool use_global_stats,
+                        bool trainable_statistics,
+                        SparseCooTensor* y,
+                        DenseTensor* mean_out,
+                        DenseTensor* variance_out,
+                        DenseTensor* saved_mean,
+                        DenseTensor* saved_variance,
+                        DenseTensor* reserve_space);
 
 }  // namespace sparse
 }  // namespace phi
diff --git a/python/paddle/sparse/nn/layer/norm.py b/python/paddle/sparse/nn/layer/norm.py
index 5714514b61355c815b07e50abcda16c5692e3119..eb242a72c54b610a29ba7c660c7af03fdd4aebd2 100644
--- a/python/paddle/sparse/nn/layer/norm.py
+++ b/python/paddle/sparse/nn/layer/norm.py
@@ -138,7 +138,7 @@ class BatchNorm(paddle.nn.BatchNorm1D):
         data_format = 'NCHW' if self._data_format[1] == 'C' else 'NHWC'
 
         if in_dynamic_mode():
-            batch_norm_out, _, _, _, _, _ = _C_ops.sparse_batch_norm(
+            batch_norm_out, _, _, _, _, _ = _C_ops.sparse_batch_norm_(
                 input,
                 self._mean,
                 self._variance,