@@ -0,0 +190,4 @@
+
+
+class TransformerDecoderLayer(Module):
+    __constants__ = ['batch_first', 'norm_first']

@@ -103,6 +103,9 @@ all = [
     'fold',
     'unfold',
 
+    'multi_head_attention_forward'

@@ -0,0 +78,4 @@
+    def generate_square_subsequent_mask(sz):
+        #TODO: replace with ms.ops.triu and ms.ops.full
+        # does not support ascend now
+        return ms.numpy.full((sz, sz), float('-inf')).triu(diagonal=1)

@@ -0,0 +275,4 @@
+
+
+def _get_clones(module, N):
+    #TODO: CellList?

@@ -9,6 +9,7 @@ from mindspore import context
 import mindspore as ms
 import torch
 import pytest
+import pytest

@@ -104,2 +104,4 @@
     'unfold',
 
+    'multi_head_attention_forward'
+    'unfold',