diff --git a/paddle/fluid/operators/fused/fused_bn_activation_op.cc b/paddle/fluid/operators/fused/fused_bn_activation_op.cc
index 97cd4d90be689ac7e891af9fe098b56bea000166..e9ad2895e03db8e77470c490453427a41d8e3bba 100644
--- a/paddle/fluid/operators/fused/fused_bn_activation_op.cc
+++ b/paddle/fluid/operators/fused/fused_bn_activation_op.cc
@@ -173,7 +173,9 @@ void FusedBatchNormActOpMaker::Make() {
       .AddCustomChecker([](const float &epsilon) {
         PADDLE_ENFORCE_EQ(epsilon >= 0.0f && epsilon <= 0.001f, true,
                           platform::errors::InvalidArgument(
-                              "'epsilon' should be between 0.0 and 0.001."));
+                              "Attr(epsilon) should be between 0.0 and 0.001, "
+                              "but received value is %f.",
+                              epsilon));
       });
   AddAttr<std::string>("act_type", "The activation type to be fused.")
       .SetDefault("relu");
diff --git a/paddle/fluid/operators/fused/fused_embedding_eltwise_layernorm_op.cc b/paddle/fluid/operators/fused/fused_embedding_eltwise_layernorm_op.cc
index b53b407d4995da5d548a13fec20ff3b09a5583c4..4d270280d389c6d8c34e3a5691a41a684b537577 100644
--- a/paddle/fluid/operators/fused/fused_embedding_eltwise_layernorm_op.cc
+++ b/paddle/fluid/operators/fused/fused_embedding_eltwise_layernorm_op.cc
@@ -25,11 +25,13 @@ class EmbeddingEltWiseLayerNormOp : public framework::OperatorWithKernel {
 
  protected:
   void InferShape(framework::InferShapeContext* context) const override {
-    PADDLE_ENFORCE_EQ(context->Inputs("Ids").size(),
-                      context->Inputs("Embs").size(),
-                      platform::errors::InvalidArgument(
-                          "Two inputs of EmbeddingEltWiseLayerNormOp shoube be "
-                          "the same size"));
+    PADDLE_ENFORCE_EQ(
+        context->Inputs("Ids").size(), context->Inputs("Embs").size(),
+        platform::errors::InvalidArgument(
+            "Two inputs of EmbeddingEltWiseLayerNormOp shoube be "
+            "the same size, but received the size of input Ids = %d,"
+            " the size of input Embs = %d",
+            context->Inputs("Ids").size(), context->Inputs("Embs").size()));
     PADDLE_ENFORCE_GE(context->Inputs("Embs").size(), 2UL,
                       platform::errors::InvalidArgument(
                           "Input Embs of EmbeddingEltWiseLayerNormOp should "
@@ -77,7 +79,8 @@ class EmbeddingEltWiseLayerNormOp : public framework::OperatorWithKernel {
       PADDLE_ENFORCE_EQ(
           embs_dims[i][1], hidden,
           platform::errors::InvalidArgument(
-              "The Emb first dim size(%d) shoule equal to hidden (%d).",
+              "The second dimension size(%d) of the Embedding should be "
+              "equal to the hidden's size(%d)",
               embs_dims[i][1], hidden));
     }
 
diff --git a/paddle/fluid/operators/fused/fusion_transpose_flatten_concat_op.cc b/paddle/fluid/operators/fused/fusion_transpose_flatten_concat_op.cc
index bd376b1e7aaefbf890e174cc86899b990a9fed26..382d01f6a535c76bdd38102a0cb40e5afc345f07 100644
--- a/paddle/fluid/operators/fused/fusion_transpose_flatten_concat_op.cc
+++ b/paddle/fluid/operators/fused/fusion_transpose_flatten_concat_op.cc
@@ -40,7 +40,9 @@ class TransposeFlattenConcatFusionOp : public framework::OperatorWithKernel {
     const size_t n = ins.size();
     PADDLE_ENFORCE_GT(n, 0,
                       platform::errors::InvalidArgument(
-                          "Input tensors dim size should greater than 0."));
+                          "The size of Inputs(X)'s dimension should be greater "
+                          " than 0, but received %d.",
+                          n));
 
     std::vector<int> trans_axis =
         ctx->Attrs().Get<std::vector<int>>("trans_axis");