add changelog and a comment

Author: dakshvar22

changelog/5529.improvement.rst

@@ -0,0 +1 @@
+Fix memory leak problem on increasing number of calls to ``/model/parse`` endpoint.

rasa/utils/tensorflow/models.py

@@ -206,7 +206,8 @@ def predict(self, predict_data: RasaModelData) -> Dict[Text, tf.Tensor]:
             logger.debug("There is no tensorflow prediction graph.")
             self.build_for_predict(predict_data)
 
-        batch_in = predict_data.prepare_batch(predict_data.data, 0, 1)
+        # Prepare a single batch of size 1
+        batch_in = predict_data.prepare_batch(start=0, end=1)
 
         self._training = False  # needed for eager mode
         return self._predict_function(batch_in)