reviese bugs

tensor2tensor/models/transformer.py

@@ -194,6 +194,9 @@ def transformer_encoder(encoder_input,
                                                       broadcast_dims=residual_dropout_broadcast_dims)
         x = residual + x
         x = may_be_layernorm(x, hparams, after=True)
+
+    if hparams.normalize_before:
+      x = may_be_layernorm(x, hparams, before=True, name="norm_top")
   return x