diff --git a/docs/source/examples/example_model.py b/docs/source/examples/example_model.py
index 5bf1ead73d..17c90cc81f 100644
--- a/docs/source/examples/example_model.py
+++ b/docs/source/examples/example_model.py
@@ -75,7 +75,8 @@ class ExampleModel(RootModule):
         # return loss_val, tqdm_dic
 
         output = OrderedDict({
-            'tng_loss': loss_val,
+            'loss': loss_val,
+            'tqdm_metrics': {}
         })
         return output
 
diff --git a/pytorch_lightning/models/trainer.py b/pytorch_lightning/models/trainer.py
index c18ae6e548..ecaf2f22e1 100644
--- a/pytorch_lightning/models/trainer.py
+++ b/pytorch_lightning/models/trainer.py
@@ -394,14 +394,17 @@ class Trainer(TrainerIO):
 
         # forward pass
         # return a scalar value and a dic with tqdm metrics
-        pdb.set_trace()
-        if self.data_parallel:
-            output = self.model(data_batch, batch_nb)
-            output = reduce_distributed_output(output, len(self.data_parallel_device_ids))
-        else:
-            output = self.mode(data_batch, batch_nb)
+        output = self.model(data_batch, batch_nb, False)
+
+        # when DP, we need to aggregate the scalars we received as outputs
+        # use mean as the reduce function
+        if self.data_parallel:
+            output = reduce_distributed_output(output, len(self.gpus))
+
+        pdb.set_trace()
+        model_specific_tqdm_metrics_dic = output['tqdm_metrics']
+        loss = output['loss']
 
-        loss, model_specific_tqdm_metrics_dic = self.model.training_step(data_batch, batch_nb)
         self.__add_tqdm_metrics(model_specific_tqdm_metrics_dic)
 
         # backward pass