使用PyInstaller打包含torchaudio(CPU版)的Python脚本为exe时遇属性错误

问题概述 #

在Windows系统中，使用无GPU支持的torchaudio编写的Python脚本直接运行正常，但通过PyInstaller打包为.exe文件后运行，出现AttributeError: '_OpNamespace' 'torchaudio' object has no attribute 'cuda_version'错误。尝试安装带CUDA的torch后问题依旧，且包体积大幅增加。

复现代码 #

import torchaudio
import time

if __name__ == '__main__':
    t = torchaudio.transforms
    time.sleep(3)
    print("Success")

错误信息 #

Traceback (most recent call last):
  File "torch\_ops.py", line 501, in __getattr__
    op, overload_names = torch._C._jit_get_operation(qualified_op_name)
RuntimeError: No such operator torchaudio::cuda_version

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "test.py", line 1, in <module>
  File "<frozen importlib._bootstrap>", line 1007, in _find_and_load
  File "<frozen importlib._bootstrap>", line 986, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 680, in _load_unlocked
  File "PyInstaller\loader\pyimod02_importers.py", line 499, in exec_module
  File "torchaudio\__init__.py", line 1, in <module>
  File "<frozen importlib._bootstrap>", line 1007, in _find_and_load
  File "<frozen importlib._bootstrap>", line 986, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 680, in _load_unlocked
  File "PyInstaller\loader\pyimod02_importers.py", line 499, in exec_module
  File "torchaudio\_extension.py", line 136, in <module>
  File "torchaudio\_extension.py", line 121, in _check_cuda_version
  File "torch\_ops.py", line 505, in __getattr__
    raise AttributeError(
AttributeError: '_OpNamespace' 'torchaudio' object has no attribute 'cuda_version'
[11648] Failed to execute script 'test' due to unhandled exception!

环境版本 #

• python==3.9.15
• torch==1.13.0（无GPU支持）
• torchaudio==0.13.0
• pyinstaller==5.6.2
• six==1.15.0
• numpy==1.22.4
• scipy==1.6.0
• sounddevice==0.4.5

解决方法 #

方法1：强制收集torchaudio所有文件 #

使用PyInstaller的--collect-all参数，确保打包时收集torchaudio的所有依赖文件，避免遗漏必要的扩展组件：

pyinstaller --collect-all torchaudio test.py

方法2：禁用CUDA检查（无GPU场景适用） #

在脚本开头添加环境变量，强制torchaudio跳过CUDA版本检查，避免触发不存在的cuda_version操作符：

import os
# 禁用CUDA检查
os.environ['TORCH_AUDIO_NO_CUDA'] = '1'

import torchaudio
import time

if __name__ == '__main__':
    t = torchaudio.transforms
    time.sleep(3)
    print("Success")

修改后再用PyInstaller打包即可。

方法3：修改.spec文件（进阶） #

如果上述方法无效，可手动编辑生成的test.spec文件，添加torchaudio的隐藏导入和二进制文件收集：

1. 生成.spec文件：pyinstaller --specpath . test.py
2. 打开test.spec，修改Analysis部分：

a = Analysis(
    ['test.py'],
    pathex=[],
    binaries=[],
    datas=[],
    hiddenimports=['torchaudio._extension', 'torchaudio._torchaudio'],
    hookspath=[],
    hooksconfig={},
    runtime_hooks=[],
    excludes=[],
    win_no_prefer_redirects=False,
    win_private_assemblies=False,
    cipher=block_cipher,
    noarchive=False,
)

3. 重新打包：pyinstaller test.spec

内容的提问来源于stack exchange，提问作者ronkov