Trance is useless

[XZRRRRRR]

Recently, when I deployed a Detectron2 model in C++, I tried using the export_tracing function in tools/deploy/export_model.py, and it works fine but the extracted model is only available for case inputs, not for other image inputs (but it is correctly available when it should); When I try to use the export_scripting function, I get an error for modeling/backbone/swin_transformer.py:
RuntimeError:
Arguments for call are not valid.
The following variants are available:

aten::view(Tensor(a) self, SymInt[] size) -> Tensor(a):
Expected a value of type 'int' for argument '' but instead found type 'NoneType'.

The original call is:
File "/home/xiangzhaorui/python_projects/SwinTextSpotter/detectron2/modeling/backbone/swin_transformer.py", line 205
shortcut = x
x = self.norm1(x)
x = x.view(B, H, W, C)
~~~~~~ <--- HERE

    # pad feature maps to multiples of window size

Is anyone can help me there? Please!!!!!!!!!