minor fixes

megatron/model/language_model.py

@@ -595,7 +595,7 @@ class TransformerLanguageModelLastStage(TransformerLanguageModelBase):
         return super(TransformerLanguageModelLastStage, self).forward(
             hidden_states,
             enc_attention_mask,
-            dec_language_input=(dec_input_ids, dec_position_ids),
+            dec_language_model_input=(dec_input_ids, dec_position_ids),
             dec_attn_mask=dec_attn_mask,
             enc_dec_attn_mask=enc_dec_attn_mask,
             layer_past=layer_past,

megatron/model/transformer.py

@@ -673,6 +673,9 @@ class ParallelTransformer(MegatronModule):
             else:
                 hidden_states = hidden_states.transpose(0, 1).contiguous()
 
+        if encoder_output is not None:
+             encoder_output = encoder_output.transpose(0, 1).contiguous()
+          
         if self.checkpoint_activations:
             hidden_states = self._checkpointed_forward(hidden_states,
                                                        attention_mask,