diff --git a/paddle/fluid/operators/CMakeLists.txt b/paddle/fluid/operators/CMakeLists.txt
index 618d31098563b..6fe18f2479478 100644
--- a/paddle/fluid/operators/CMakeLists.txt
+++ b/paddle/fluid/operators/CMakeLists.txt
@@ -187,6 +187,4 @@ endif()
 
 if(WITH_ASCEND_CL)
 cc_test(gelu_op_npu_test SRCS gelu_op_npu_test.cc DEPS op_registry gelu_op scope device_context enforce executor)
-cc_test(mean_op_npu_test SRCS mean_op_npu_test.cc DEPS op_registry mean_op scope device_context enforce executor)
 endif()
-
diff --git a/paddle/fluid/operators/lookup_table_v2_op_npu.cc b/paddle/fluid/operators/lookup_table_v2_op_npu.cc
index e7cc048ed3ce4..8ab4d70fd3ff6 100644
--- a/paddle/fluid/operators/lookup_table_v2_op_npu.cc
+++ b/paddle/fluid/operators/lookup_table_v2_op_npu.cc
@@ -54,6 +54,7 @@ class LookupTableV2GradNPUKernel : public framework::OpKernel<T> {
     auto *table_t = ctx.Input<framework::LoDTensor>("W");
     auto *table_grad_t =
         ctx.Output<framework::LoDTensor>(framework::GradVarName("W"));
+    table_grad_t->mutable_data<T>(ctx.GetPlace());
     framework::NPUAttributeMap attr_input = {{"use_locking", true}};
 
     auto runner = NpuOpRunner("ScatterAdd", {*table_t, *ids_t, *output_grad_t},
diff --git a/paddle/fluid/operators/softmax_op.cc b/paddle/fluid/operators/softmax_op.cc
index 7d4194227b4cd..a38800da87fd1 100644
--- a/paddle/fluid/operators/softmax_op.cc
+++ b/paddle/fluid/operators/softmax_op.cc
@@ -206,9 +206,10 @@ class SoftmaxOpGrad : public framework::OperatorWithKernel {
     auto input_data_type = OperatorWithKernel::IndicateVarDataType(
         ctx, framework::GradVarName("Out"));
     if (input_data_type == framework::proto::VarType::FP16) {
-      PADDLE_ENFORCE_EQ(platform::is_gpu_place(ctx.GetPlace()), true,
-                        platform::errors::InvalidArgument(
-                            "float16 can only be used on GPU place"));
+      if (!(platform::is_gpu_place(ctx.GetPlace()) ||
+            platform::is_npu_place(ctx.GetPlace())))
+        PADDLE_THROW(platform::errors::InvalidArgument(
+            "float16 can only be used on GPU/NPU place"));
     }
 
     return framework::OpKernelType(input_data_type, ctx.GetPlace(), layout_,