fix m,n class on OrangePi

mindtorch/_apis/npu.py

@@ -1887,7 +1887,7 @@ def relu6(input):
 def col2im(input, output_size, kernel_size, dilation=1, padding=0, stride=1):
     if use_pyboost():
         return pyboost.col2im_ext_op(input, output_size, kernel_size, dilation, padding, stride)
-    return legacy.col2im(input, output_size, kernel_size, dilation, padding, stride)
+    return legacy.col2_im(input, mindspore.Tensor(output_size), kernel_size, dilation, padding, stride)
 
 def flash_attention_score(query, key, value, real_shift, drop_mask, padding_mask, attn_mask, prefix, actual_seq_qlen, actual_seq_kvlen, head_num, keep_prob, scale_value, pre_tokens, next_tokens, inner_precise, input_layout, sparse_mode):
     if use_pyboost():
@@ -1992,4 +1992,10 @@ def logaddexp(input, other):
     return y
 
 def reflection_pad_1d(input, padding):
-    return pyboost.reflection_pad_1d_op(input, padding)
+    return pyboost.reflection_pad_1d_op(input, padding)
+
+def replication_pad_1d(input, padding):
+    return pyboost.reflection_pad_1d_op(input, padding)
+
+def hardtanh(input, min_val, max_val):
+    return pyboost.hardtanh_op(input, min_val, max_val)

mindtorch/nn/functional.py

@@ -1897,4 +1897,7 @@ def make_causal_mask(
     )
 
 def rotary_position_embedding(x, cos, sin, mode=0):
-    return ops.rotary_position_embedding(x, cos, sin, mode)
+    return ops.rotary_position_embedding(x, cos, sin, mode)
+
+def hardtanh(input, min_val=-1.0, max_val=1.0):
+    return execute('hardtanh', input, min_val, max_val)

mindtorch/nn/modules/activation.py

@@ -289,7 +289,7 @@ def forward(self, input: Tensor) -> Tensor:
         """
         Runs the forward pass.
         """
-        return F.hardtanh(input, self.min_val, self.max_val, self.inplace)
+        return F.hardtanh(input, self.min_val, self.max_val)
 
     def extra_repr(self) -> str:
         """

mindtorch/nn/modules/dropout.py

@@ -173,7 +173,7 @@ def forward(self, input: Tensor) -> Tensor:
         """
         Runs the forward pass.
         """
-        return F.dropout2d(input, self.p, self.training, self.inplace)
+        return F.dropout2d(input, self.p, self.training)
 
 
 class Dropout3d(_DropoutNd):