DirectML run fail on some cases.

[DeruVN]

Hi
DirectML run well sometimes, but on some performance it give errors.
(Always error with some images input, some other is always ok. Beside that all test images works fine with nvidia vga)

Another strange thing. After several errors, I can't do "import torch" in python.exe again, have to re-install torch. Maybe directml changed something in lib.

Please take a look.

Traceback (most recent call last):
File "D:\anaconda3\Depth-Anything-V2\run.py", line 60, in
depth = depth_anything.infer_image(raw_image, args.input_size)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\anaconda3\Lib\site-packages\torch\utils_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "D:\anaconda3\Depth-Anything-V2\depth_anything_v2\dpt.py", line 191, in infer_image
depth = self.forward(image)
^^^^^^^^^^^^^^^^^^^
File "D:\anaconda3\Depth-Anything-V2\depth_anything_v2\dpt.py", line 180, in forward
features = self.pretrained.get_intermediate_layers(x, self.intermediate_layer_idx[self.encoder], return_class_token=True)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\anaconda3\Depth-Anything-V2\depth_anything_v2\dinov2.py", line 309, in get_intermediate_layers
outputs = self._get_intermediate_layers_not_chunked(x, n)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\anaconda3\Depth-Anything-V2\depth_anything_v2\dinov2.py", line 278, in _get_intermediate_layers_not_chunked
x = blk(x)
^^^^^^
File "D:\anaconda3\Lib\site-packages\torch\nn\modules\module.py", line 1532, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\anaconda3\Lib\site-packages\torch\nn\modules\module.py", line 1541, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\anaconda3\Depth-Anything-V2\depth_anything_v2\dinov2_layers\block.py", line 248, in forward
return super().forward(x_or_x_list)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\anaconda3\Depth-Anything-V2\depth_anything_v2\dinov2_layers\block.py", line 106, in forward
x = x + attn_residual_func(x)
^^^^^^^^^^^^^^^^^^^^^
File "D:\anaconda3\Depth-Anything-V2\depth_anything_v2\dinov2_layers\block.py", line 85, in attn_residual_func
return self.ls1(self.attn(self.norm1(x)))
^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\anaconda3\Lib\site-packages\torch\nn\modules\module.py", line 1532, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\anaconda3\Lib\site-packages\torch\nn\modules\module.py", line 1541, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "D:\anaconda3\Depth-Anything-V2\depth_anything_v2\dinov2_layers\attention.py", line 69, in forward
return super().forward(x)
^^^^^^^^^^^^^^^^^^
File "D:\anaconda3\Depth-Anything-V2\depth_anything_v2\dinov2_layers\attention.py", line 54, in forward
attn = q @ k.transpose(-2, -1)
~~^~~~~~~~~~~~~~~~~~~~~
RuntimeError: The parameter is incorrect.