fix a bug of transformertts when use data parallel.

examples/transformer_tts/train_transformer.py

@@ -94,10 +94,16 @@ def main(args):
                     if args.stop_token:
                         writer.add_scalar('stop_loss', stop_loss.numpy(), global_step)
 
-                    writer.add_scalars('alphas', {
-                        'encoder_alpha':model.encoder.alpha.numpy(),
-                        'decoder_alpha':model.decoder.alpha.numpy(),
-                    }, global_step)
+                    if args.use_data_parallel:
+                        writer.add_scalars('alphas', {
+                            'encoder_alpha':model._layers.encoder.alpha.numpy(),
+                            'decoder_alpha':model._layers.decoder.alpha.numpy(),
+                        }, global_step)
+                    else:
+                        writer.add_scalars('alphas', {
+                            'encoder_alpha':model.encoder.alpha.numpy(),
+                            'decoder_alpha':model.decoder.alpha.numpy(),
+                        }, global_step)
 
                     writer.add_scalar('learning_rate', optimizer._learning_rate.step().numpy(), global_step)
 
@@ -144,4 +150,4 @@ if __name__ =='__main__':
     args = parser.parse_args()
     # Print the whole config setting.
     pprint(args)
-    main(args)
\ No newline at end of file
+    main(args)