diff --git a/paddle/fluid/inference/api/helper.h b/paddle/fluid/inference/api/helper.h
index 24f59cf43a9700ff1732e1ef6ad82e1a6294eede..e46dc1326951f68fd030f2208b9bea1647d0026d 100644
--- a/paddle/fluid/inference/api/helper.h
+++ b/paddle/fluid/inference/api/helper.h
@@ -160,7 +160,8 @@ static void PrintTime(int batch_size, int repeat, int num_threads, int tid,
                       double latency, int epoch = 1) {
   LOG(INFO) << "====== batch_size: " << batch_size << ", repeat: " << repeat
             << ", threads: " << num_threads << ", thread id: " << tid
-            << ", latency: " << latency << "ms ======";
+            << ", latency: " << latency << "ms, fps: " << 1 / (latency / 1000.f)
+            << " ======";
   if (epoch > 1) {
     int samples = batch_size * epoch;
     LOG(INFO) << "====== sample number: " << samples
diff --git a/paddle/fluid/inference/tests/api/tester_helper.h b/paddle/fluid/inference/tests/api/tester_helper.h
index 5589b58b0618523b22f169f9ba9930c3ff3e3c48..42072895fc66b2765570cc9a87809ab796057749 100644
--- a/paddle/fluid/inference/tests/api/tester_helper.h
+++ b/paddle/fluid/inference/tests/api/tester_helper.h
@@ -139,6 +139,7 @@ void TestMultiThreadPrediction(
   }
   for (int tid = 0; tid < num_threads; ++tid) {
     threads.emplace_back([&, tid]() {
+      platform::set_cur_thread_id(static_cast<int>(tid) + 1);
       // Each thread should have local inputs and outputs.
       // The inputs of each thread are all the same.
       std::vector<std::vector<PaddleTensor>> inputs_tid = inputs;