Bypassing Export model step, if training on TPU's. As this need inference to...

Bypassing Export model step, if training on TPU's. As this need inference to be supported on TPU's. Remove this check once inference is supported.

official/transformer/transformer_main.py

@@ -610,7 +610,7 @@ def run_transformer(flags_obj):
       bleu_threshold=flags_obj.stop_threshold,
       vocab_file=flags_obj.vocab_file)
 
-  if flags_obj.export_dir:
+  if flags_obj.export_dir and not params["use_tpu"]:
     serving_input_fn = export.build_tensor_serving_input_receiver_fn(
         shape=[None], dtype=tf.int64, batch_size=None)
     # Export saved model, and save the vocab file as an extra asset. The vocab