diff --git a/paddle/phi/kernels/gpu/sigmoid_cross_entropy_with_logits_grad_kernel.cu b/paddle/phi/kernels/gpu/sigmoid_cross_entropy_with_logits_grad_kernel.cu
index 6fc65006ae264e29a5637cb18a95b5785c922f07..f61cd2c39674ecdad5553e8472fb85da73ebd534 100644
--- a/paddle/phi/kernels/gpu/sigmoid_cross_entropy_with_logits_grad_kernel.cu
+++ b/paddle/phi/kernels/gpu/sigmoid_cross_entropy_with_logits_grad_kernel.cu
@@ -95,6 +95,7 @@ void SigmoidCrossEntropyWithLogitsGradKernel(const Context &dev_ctx,
                          norm,
                          sizeof(T),
                          dev_ctx.stream());
+    dev_ctx.Wait();
     auto eps = static_cast<T>(1e-5);
     *norm_cpu_ptr = *norm_cpu_ptr > eps ? *norm_cpu_ptr : eps;
 
@@ -102,6 +103,7 @@ void SigmoidCrossEntropyWithLogitsGradKernel(const Context &dev_ctx,
     std::vector<DenseTensor *> div_outs = {in_grad};
     auto div_functor = DivFunctor<T>(*norm_cpu_ptr);
     phi::funcs::ElementwiseKernel<T>(dev_ctx, div_ins, &div_outs, div_functor);
+
     delete norm_tensor;
   }
   delete counts_tensor;
diff --git a/paddle/phi/kernels/gpu/sigmoid_cross_entropy_with_logits_kernel.cu b/paddle/phi/kernels/gpu/sigmoid_cross_entropy_with_logits_kernel.cu
index 4b6e5628c72af438fc98cf6f64c5a36adcd6d2ee..b0e9efe5bbafe671ae6f594564260895a2015a45 100644
--- a/paddle/phi/kernels/gpu/sigmoid_cross_entropy_with_logits_kernel.cu
+++ b/paddle/phi/kernels/gpu/sigmoid_cross_entropy_with_logits_kernel.cu
@@ -95,6 +95,7 @@ void SigmoidCrossEntropyWithLogitsKernel(const Context &dev_ctx,
                          norm,
                          sizeof(T),
                          dev_ctx.stream());
+    dev_ctx.Wait();
     auto eps = static_cast<T>(1e-5);
     *norm_cpu_ptr = *norm_cpu_ptr > eps ? *norm_cpu_ptr : eps;