diff --git a/python/examples/bert/benchmark.py b/python/examples/bert/benchmark.py
index eff82051c37dcbfef6002e2e14af643dac4864a9..bdef982830cea34f5a9ea925e6759b48b86ce7a7 100755
--- a/python/examples/bert/benchmark.py
+++ b/python/examples/bert/benchmark.py
@@ -70,7 +70,8 @@ def single_func(idx, resource):
                             os.getpid(),
                             int(round(b_start * 1000000)),
                             int(round(b_end * 1000000))))
-                result = client.predict(feed=feed_batch, fetch=fetch)
+                result = client.predict(
+                    feed=feed_batch, fetch=fetch, batch=True)
 
                 l_end = time.time()
                 if latency_flags: