!8984 Fix evaluation error in data parallel mode and standalone if sparse is set

From: @huangxinjing
Reviewed-by: @yao_yf,@stsuteng,@chenfei52
Signed-off-by: @stsuteng

model_zoo/official/recommend/wide_and_deep/train_and_eval.py

@@ -95,7 +95,7 @@ def test_train_eval(config):
     ckptconfig = CheckpointConfig(save_checkpoint_steps=ds_train.get_dataset_size(), keep_checkpoint_max=5)
     ckpoint_cb = ModelCheckpoint(prefix='widedeep_train', directory=config.ckpt_path, config=ckptconfig)
 
-    out = model.eval(ds_eval)
+    out = model.eval(ds_eval, dataset_sink_mode=(not sparse))
     print("=====" * 5 + "model.eval() initialized: {}".format(out))
     model.train(epochs, ds_train,
                 callbacks=[TimeMonitor(ds_train.get_dataset_size()), eval_callback, callback, ckpoint_cb],

model_zoo/official/recommend/wide_and_deep/train_and_eval_distribute.py

@@ -105,7 +105,7 @@ def train_and_eval(config):
     ckpoint_cb = ModelCheckpoint(prefix='widedeep_train',
                                  directory=config.ckpt_path + '/ckpt_' + str(get_rank()) + '/',
                                  config=ckptconfig)
-    out = model.eval(ds_eval)
+    out = model.eval(ds_eval, dataset_sink_mode=(not sparse))
     print("=====" * 5 + "model.eval() initialized: {}".format(out))
     callback_list = [TimeMonitor(ds_train.get_dataset_size()), eval_callback, callback]
     if get_rank() == 0: