AttributeError: type object 'torch._C.ErrorReport' has no attribute 'call_stack'

I use the pytorch environment(torch=1.4.0+CU9.2) to train the ASCON model, but when import torch, I get the following error1, which I think is because my torch version is too low, so I copy part of the package from torch1.10.0 into my torch1.4.0 environment to replace it, but the following error2 still occurs:

“C:\Program Files\Anaconda3\envs\torch\python.exe” C:/Users/Administrator/PycharmProjects/ASCON/
Traceback (most recent call last):
File “C:/Users/Administrator/PycharmProjects/ASCON/”, line 2, in
import torch
File “C:\Program Files\Anaconda3\envs\torch\lib\site-packages\”, line 280, in
from .functional import *
File “C:\Program Files\Anaconda3\envs\torch\lib\site-packages\torch\”, line 2, in
import torch.nn.functional as F
File “C:\Program Files\Anaconda3\envs\torch\lib\site-packages\torch\”, line 1, in
from .modules import *
File “C:\Program Files\Anaconda3\envs\torch\lib\site-packages\torch\nn\”, line 23, in
from .rnn import RNNBase, RNN, LSTM, GRU,
File “C:\Program Files\Anaconda3\envs\torch\lib\site-packages\torch\nn\modules\”, line 399, in
class LSTM(RNNBase):
File “C:\Program Files\Anaconda3\envs\torch\lib\site-packages\torch\nn\modules\”, line 523, in LSTM
def forward(self, input, hx=None): # noqa: F811
File “C:\Program Files\Anaconda3\envs\torch\lib\site-packages\”, line 814, in _overload_method
File “C:\Program Files\Anaconda3\envs\torch\lib\site-packages\”, line 746, in _check_overload_body
parsed_def = parse_def(func)
File “C:\Program Files\Anaconda3\envs\torch\lib\site-packages\”, line 95, in parse_def
sourcelines, file_lineno, filename = get_source_lines_and_file(fn, ErrorReport.call_stack())
AttributeError: type object ‘torch._C.ErrorReport’ has no attribute ‘call_stack’

Due to the hardware limitations of my computer, CUDA can only be installed with version 9.2 and below, so there is no way to directly upgrade the torch environment, I would appreciate it if you could answer my questions

This doesn’t sound like a good idea as it could create all kinds of conflicts.
Would it possible to update PyTorch instead?