!15338 fix dataset sink hang up in PyNative mode

From: @chujinjin
Reviewed-by: @zhoufeng54,@jjfeing
Signed-off-by: @jjfeing

mindspore/train/model.py

@@ -438,18 +438,23 @@ class Model:
             cb_params (_InternalCallbackParam): Callback parameters. Default: None.
             sink_size (int): Control the amount of data in each sink. Default: -1.
         """
+        is_graph = (context.get_context("mode") == context.GRAPH_MODE)
         if sink_size == -1:
             epoch_num = epoch
         else:
-            epoch_num = math.ceil(epoch * sink_size / train_dataset.get_dataset_size())
-            train_dataset.__total_batch__ = epoch * sink_size
+            if is_graph:
+                epoch_num = math.ceil(epoch * sink_size / train_dataset.get_dataset_size())
+                train_dataset.__total_batch__ = epoch * sink_size
+            else:
+                sink_size = -1
+                epoch_num = epoch
+                logger.warning("Loop sink is not supported in PyNative mode, so it will be performed with no loop sink")
 
         cb_params.cur_step_num = 0
         cb_params.dataset_sink_mode = True
 
         run_context = RunContext(cb_params)
         list_callback.begin(run_context)
-        is_graph = (context.get_context("mode") == context.GRAPH_MODE)
         # used to stop training for early stop, such as stopAtTIme or stopATStep
         should_stop = False
         dataset_helper = None