diff --git a/examples/text/information_extraction/ernie/cpp/uie.cc b/examples/text/information_extraction/ernie/cpp/uie.cc
index 93c5adfeee..0e5a02f5f8 100644
--- a/examples/text/information_extraction/ernie/cpp/uie.cc
+++ b/examples/text/information_extraction/ernie/cpp/uie.cc
@@ -75,7 +75,7 @@ Schema::Schema(
 
 UIEModel::UIEModel(const std::string& model_file,
                    const std::string& params_file,
-                   const std::string& vocab_file, double position_prob,
+                   const std::string& vocab_file, float position_prob,
                    size_t max_length, const std::vector<std::string>& schema)
     : max_length_(max_length),
       position_prob_(position_prob),
@@ -90,7 +90,7 @@ UIEModel::UIEModel(const std::string& model_file,
 
 UIEModel::UIEModel(
     const std::string& model_file, const std::string& params_file,
-    const std::string& vocab_file, double position_prob, size_t max_length,
+    const std::string& vocab_file, float position_prob, size_t max_length,
     const std::unordered_map<std::string, std::vector<std::string>>& schema)
     : max_length_(max_length),
       position_prob_(position_prob),
@@ -155,6 +155,20 @@ void UIEModel::AutoSplitter(
   }
 }
 
+void UIEModel::GetCandidateIdx(
+    const float* probs, int64_t batch_size, int64_t seq_len,
+    std::vector<std::vector<std::pair<int64_t, float>>>* candidate_idx_prob,
+    float threshold) const {
+  for (int i = 0; i < batch_size; ++i) {
+    candidate_idx_prob->push_back({});
+    for (int j = 0; j < seq_len; ++j) {
+      if (probs[i * seq_len + j] > threshold) {
+        candidate_idx_prob->back().push_back({j, probs[i * seq_len + j]});
+      }
+    }
+  }
+}
+
 void UIEModel::PredictUIEInput(const std::vector<std::string>& input_texts,
                                const std::vector<std::string>& prompts) {
   // 1. Shortten the input texts and prompts
@@ -214,17 +228,20 @@ void UIEModel::PredictUIEInput(const std::vector<std::string>& input_texts,
     inputs[i].name = runtime_.GetInputInfo(i).name;
   }
 
-  std::vector<float> start_probs, end_probs;
   std::vector<fastdeploy::FDTensor> outputs(runtime_.NumOutputs());
   // 4. Infer
   runtime_.Infer(inputs, &outputs);
   auto* start_prob = reinterpret_cast<float*>(outputs[0].Data());
   auto* end_prob = reinterpret_cast<float*>(outputs[1].Data());
-  start_probs.insert(start_probs.end(), start_prob,
-                     start_prob + outputs[0].Numel());
-  end_probs.insert(end_probs.end(), end_prob, end_prob + outputs[0].Numel());
 
   // 5. Postprocess
+  std::vector<std::vector<std::pair<int64_t, float>>> start_candidate_idx_prob,
+      end_candidate_idx_prob;
+
+  GetCandidateIdx(start_prob, outputs[0].shape[0], outputs[0].shape[1],
+                  &start_candidate_idx_prob, position_prob_);
+  GetCandidateIdx(end_prob, outputs[1].shape[0], outputs[1].shape[1],
+                  &end_candidate_idx_prob, position_prob_);
 }
 
 void UIEModel::Predict(const std::vector<std::string>& texts,
diff --git a/examples/text/information_extraction/ernie/cpp/uie.h b/examples/text/information_extraction/ernie/cpp/uie.h
index 583aeeb25a..a638181246 100644
--- a/examples/text/information_extraction/ernie/cpp/uie.h
+++ b/examples/text/information_extraction/ernie/cpp/uie.h
@@ -70,11 +70,11 @@ struct UIEInput {
 
 struct UIEModel {
   UIEModel(const std::string& model_file, const std::string& params_file,
-           const std::string& vocab_file, double position_prob,
+           const std::string& vocab_file, float position_prob,
            size_t max_length, const std::vector<std::string>& schema);
   UIEModel(
       const std::string& model_file, const std::string& params_file,
-      const std::string& vocab_file, double position_prob, size_t max_length,
+      const std::string& vocab_file, float position_prob, size_t max_length,
       const std::unordered_map<std::string, std::vector<std::string>>& schema);
   void SetSchema(const std::vector<std::string>& schema);
   void SetSchema(
@@ -90,10 +90,16 @@ struct UIEModel {
       const std::vector<std::string>& texts, size_t max_length,
       std::vector<std::string>* short_texts,
       std::unordered_map<size_t, std::vector<size_t>>* input_mapping);
+  // Get idx of the last dimension in probability arrays, which is greater than
+  // a limitation.
+  void GetCandidateIdx(
+      const float* probs, int64_t batch_size, int64_t seq_len,
+      std::vector<std::vector<std::pair<int64_t, float>>>* candidate_idx_prob,
+      float threshold = 0.5) const;
   fastdeploy::RuntimeOption runtime_option_;
   fastdeploy::Runtime runtime_;
   std::unique_ptr<Schema> schema_;
   size_t max_length_;
-  double position_prob_;
+  float position_prob_;
   faster_tokenizer::tokenizers_impl::ErnieFasterTokenizer tokenizer_;
 };