diff --git a/paddle/fluid/operators/fused/fused_seq_tensor_op.cc b/paddle/fluid/operators/fused/fused_seq_tensor_op.cc
index 5ca2ec345f10e..7430d0d32ca37 100644
--- a/paddle/fluid/operators/fused/fused_seq_tensor_op.cc
+++ b/paddle/fluid/operators/fused/fused_seq_tensor_op.cc
@@ -47,10 +47,10 @@ class FusedSeqTensorOp : public framework::OperatorWithKernel {
         ad_slot_num, 0, 
         platform::errors::InvalidArgument(
           "ad_slot_num [%ld] <= 0", ad_slot_num));
-    PADDLE_ENFORCE_LT(
+    PADDLE_ENFORCE_LE(
         ad_slot_offset, slot_num - 1, 
         platform::errors::InvalidArgument(
-          "ad_slot_num [%ld] >  slot_num - 1 [%ld]", ad_slot_offset, slot_num));
+          "ad_slot_num [%ld] > slot_num - 1 [%ld]", ad_slot_offset, slot_num));
     PADDLE_ENFORCE_GE(
         ad_slot_offset, 0, 
         platform::errors::InvalidArgument(
diff --git a/paddle/fluid/operators/fused/fused_seq_tensor_op.cu b/paddle/fluid/operators/fused/fused_seq_tensor_op.cu
index d2fdf364d731d..8210cd43808c3 100644
--- a/paddle/fluid/operators/fused/fused_seq_tensor_op.cu
+++ b/paddle/fluid/operators/fused/fused_seq_tensor_op.cu
@@ -145,8 +145,8 @@ __device__ void warpReduce(volatile T* cache, int tid) {
 
 #define THREAD_PER_BLOCK 128
 template <typename T>
-__global__ void reduce_sum_max_length(const T* input,  // 1
-                                      T* mask_output,  // mask
+__global__ void reduce_sum_max_length(const T* input,
+                                      T* mask_output,
                                       const size_t batch_count,
                                       const size_t ins_num,
                                       const size_t slot_num,