module ‘torch.jit‘ has no attribute ‘unused‘问题

在执行from torchvision import transforms时出现了以下错误：

Traceback (most recent call last):
  File "D:/研究生/毕业设计/COVID19/COVID19_CT/conduct.py", line 8, in <module>
    from torchvision import transforms
  File "D:\Anaconda3\envs\pytorch_1.2_gpu\lib\site-packages\torchvision\__init__.py", line 3, in <module>
    from torchvision import models
  File "D:\Anaconda3\envs\pytorch_1.2_gpu\lib\site-packages\torchvision\models\__init__.py", line 5, in <module>
    from .inception import *
  File "D:\Anaconda3\envs\pytorch_1.2_gpu\lib\site-packages\torchvision\models\inception.py", line 63, in <module>
    class Inception3(nn.Module):
  File "D:\Anaconda3\envs\pytorch_1.2_gpu\lib\site-packages\torchvision\models\inception.py", line 180, in Inception3
    @torch.jit.unused
AttributeError: module 'torch.jit' has no attribute 'unused'

笔者本机环境：

• windows7系统
• torch==1.2.0+cu92
• torchvision== 0.6.1+cu92

 

解决方法：

安装新版torch。原来安装的是pytorch1.2.0，需要更新为pytorch1.4.0及以上版本。

登录https://download.pytorch.org/whl/torch_stable.html，选择cu92/torch-1.4.0%2Bcu92-cp37-cp37m-win_amd64.whl（具体版本根据实际情况进行选择，笔者的环境对应这个版本） 。

 

这次的本机环境为：

• windows7系统
• torch==1.4.0+cu92
• torchvision== 0.6.1+cu92

在执行from torchvision import transforms时还是有错误：

Traceback (most recent call last):
  File "D:/研究生/毕业设计/COVID19/COVID19_CT/conduct.py", line 8, in <module>
    from torchvision import transforms
  File "D:\Anaconda3\envs\pytorch_1.4_gpu\lib\site-packages\torchvision\__init__.py", line 3, in <module>
    from torchvision import models
  File "D:\Anaconda3\envs\pytorch_1.4_gpu\lib\site-packages\torchvision\models\__init__.py", line 12, in <module>
    from . import detection
  File "D:\Anaconda3\envs\pytorch_1.4_gpu\lib\site-packages\torchvision\models\detection\__init__.py", line 1, in <module>
    from .faster_rcnn import *
  File "D:\Anaconda3\envs\pytorch_1.4_gpu\lib\site-packages\torchvision\models\detection\faster_rcnn.py", line 7, in <module>
    from torchvision.ops import misc as misc_nn_ops
  File "D:\Anaconda3\envs\pytorch_1.4_gpu\lib\site-packages\torchvision\ops\__init__.py", line 13, in <module>
    _register_custom_op()
  File "D:\Anaconda3\envs\pytorch_1.4_gpu\lib\site-packages\torchvision\ops\_register_onnx_ops.py", line 51, in _register_custom_op
    register_custom_op_symbolic('torchvision::_new_empty_tensor_op', new_empty_tensor_op, _onnx_opset_version)
  File "D:\Anaconda3\envs\pytorch_1.4_gpu\lib\site-packages\torch\onnx\__init__.py", line 200, in register_custom_op_symbolic
    return utils.register_custom_op_symbolic(symbolic_name, symbolic_fn, opset_version)
  File "D:\Anaconda3\envs\pytorch_1.4_gpu\lib\site-packages\torch\onnx\utils.py", line 793, in register_custom_op_symbolic
    .format(symbolic_name))
RuntimeError: Failed to register operator torchvision::_new_empty_tensor_op.                            The symbolic name must match the format Domain::Name,                            and sould start with a letter and contain only                            alphanumerical characters

Process finished with exit code 1

解决方法：

安装旧版torchvision。将原来安装的torchvision-0.6.1通过pip uninstall卸载，安装torchvision-0.5.0。

登录https://download.pytorch.org/whl/torch_stable.html，选择torchvision-0.5.0-cp37-cp37m-win_amd64.whl。

这次就没有问题了。