moved cuda flags inside trainer

pytorch_lightning/models/trainer.py

@@ -90,7 +90,6 @@ class Trainer(TrainerIO):
         # gpus come in as a string.
         # if gpus = -1 then use all available devices
         # otherwise, split the string using commas
-        pdb.set_trace()
         if gpus is not None:
             if gpus == '-1':
                 self.data_parallel_device_ids = list(range(0, torch.cuda.device_count()))