Cannot import torchaudio with torch 1.1

[f90]

I am trying to use torchaudio with torch 1.1.
It compiled successfully with Python 3.6, but when I want to import the torchaudio package, I get this error:

>>> import torchaudio
Traceback (most recent call last):
  File "/home/daniel/envs/pytorch-py3/lib/python3.6/site-packages/torch/jit/annotations.py", line 95, in parse_type_line
    arg_ann = eval(arg_ann_str, _eval_env)
  File "<string>", line 1, in <module>
NameError: name 'Optional' is not defined

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/daniel/envs/pytorch-py3/lib/python3.6/site-packages/torchaudio-0.2-py3.6-linux-x86_64.egg/torchaudio/__init__.py", line 7, in <module>
    from torchaudio import transforms, datasets, kaldi_io, sox_effects, legacy
  File "/home/daniel/envs/pytorch-py3/lib/python3.6/site-packages/torchaudio-0.2-py3.6-linux-x86_64.egg/torchaudio/transforms.py", line 6, in <module>
    from . import functional as F
  File "/home/daniel/envs/pytorch-py3/lib/python3.6/site-packages/torchaudio-0.2-py3.6-linux-x86_64.egg/torchaudio/functional.py", line 108, in <module>
    @torch.jit.script
  File "/home/daniel/envs/pytorch-py3/lib/python3.6/site-packages/torch/jit/__init__.py", line 824, in script
    fn = torch._C._jit_script_compile(ast, _rcb, get_default_args(obj))
  File "/home/daniel/envs/pytorch-py3/lib/python3.6/site-packages/torch/jit/annotations.py", line 55, in get_signature
    return parse_type_line(type_line)
  File "/home/daniel/envs/pytorch-py3/lib/python3.6/site-packages/torch/jit/annotations.py", line 97, in parse_type_line
    raise RuntimeError("Failed to parse the argument list of a type annotation: {}".format(str(e)))
RuntimeError: Failed to parse the argument list of a type annotation: name 'Optional' is not defined

It ran fine on torch 1.0 - maybe some problem with differing source code between torch 1.0 and 1.1?

[zqs01]

the same question.

[jamarshon]

Yes this is a known issue. We are currently working on a fix and will update this when resolved

[RJ722]

Hey, I've a dependency in my code - torch-baidu-ctc - which requires torch>1.0 - so, I can't downgrade to 1.0. What should I do?

[tobyclh]

same issue, working after I check out d92de5b

[MNCTTY]

Same issue.
I downgraded pytorch to version 1.0.0
and then installed audio as pip3 install git+https://github.com/pytorch/audio@d92de5b
it has been built successfully, but when I tried to import, I recieved this error:

I couldn't google anything (all mentions by the error was about maskrcnn, I didn't find some correspondings)

After that I tried to build audio from git as it is recommended:
git clone https://github.com/pytorch/audio.git
and then cd audio; python setup.py install
and it gives me the following traceback:

I tried to do recommendations from here - #49 - but it nothing has changed
In other issue people say that people, who meet the same error, use old versions of torch (0.3 and others), but I use 1.0.0

I use python 3.5, cuda 10.1, Ubuntu 16.04.6 LTS (Xenial Xerus)

Does someone have any ideas?

[jamarshon]

@MNCTTY For your first issue about UndefinedTensorImp, it is mentioned here https://discuss.pytorch.org/t/torchaudio-import-error/48367 where you have to upgrade your pytorch version (e.g 1.1.0)

For your second issue, you did not checkout the specific commit. You can do this by running
git clone --branch v0.2.0 https://github.com/pytorch/audio.git
or if you already cloned it you can do
git checkout d92de5b
as mentioned by tobyclh.
When it says No module named torch.utils.cpp_extension, could you try running

# >> python
import torch
print(torch.__version__)

and seeing that you can in fact import the module and that it is the correct version? I looked at v1.0.0 and it has the file torch.utils.cpp_extension.py.

Let us know if those steps still do not work for you

[mdhasanai]

Same Issue.

[thetimeofblack]

Yes this is a known issue. We are currently working on a fix and will update this when resolved
Is there any other way for me to use pytorch audio, I tries many other way to install it but it always fails

[jamarshon]

@thetimeofblack @m-d-hasan can you confirm that you have pytorch > 1.1.0 installed? e.g

# >> python
import torch
print(torch.__version__)

can you also confirm that you checkout the correct torchaudio branch/commit (v0.2.0 or d92de5b)?

If that does not work, please comment with your setup (i.e. copy and paste the output of collect_env.py, as well as the commands used to install torchaudio) so we can provide further assistance 😃

[XingwXiong]

Thanks a lot @jamarshon , I solved this error by
pip uninstall torchaudio; git checkout d92de5b; python setup.py install.

[thetimeofblack]

@jamarshon
I am confirmed the version of torch is 1.1.0

print(torch.version)
1.1.0
and
this is my collect_env.py output
Collecting environment information...
Traceback (most recent call last):
File "collect_env.py", line 386, in
main()
File "collect_env.py", line 381, in main
output = get_pretty_env_info()
File "collect_env.py", line 376, in get_pretty_env_info
return pretty_str(get_env_info())
File "collect_env.py", line 270, in get_env_info
cudnn_version=get_cudnn_version(run_lambda),
File "collect_env.py", line 130, in get_cudnn_version
rc, out, _ = run_lambda(cudnn_cmd)
File "collect_env.py", line 46, in run
err = err.decode("utf-8")
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xd0 in position 0: invalid continuation byte

I will try with python3.6.

[tobyclh]

You need to run the script and post the output, not the script itself.
it should look something like this.

Collecting environment information...
PyTorch version: 1.1.0
Is debug build: No
CUDA used to build PyTorch: 10.0.130

OS: Ubuntu 18.04.2 LTS
GCC version: (Ubuntu 7.4.0-1ubuntu1~18.04.1) 7.4.0
CMake version: version 3.10.2

Python version: 3.7
Is CUDA available: Yes
CUDA runtime version: 10.0.130
GPU models and configuration: GPU 0: GeForce GTX 1060
Nvidia driver version: 410.104
cuDNN version: Could not collect

Versions of relevant libraries:
[pip3] numpy==1.16.2
[pip3] numpydoc==0.8.0
[pip3] torch==1.1.0
[pip3] torchaudio==0.2
[pip3] torchfile==0.1.0
[pip3] torchvision==0.3.0
[conda] blas                      1.0                         mkl  
[conda] mkl                       2019.3                      199  
[conda] mkl-service               1.1.2            py37he904b0f_5  
[conda] mkl_fft                   1.0.10           py37ha843d7b_0  
[conda] mkl_random                1.0.2            py37hd81dba3_0  
[conda] pytorch                   1.1.0           py3.7_cuda10.0.130_cudnn7.5.1_0    pytorch
[conda] torchaudio                0.2                      pypi_0    pypi
[conda] torchfile                 0.1.0                    pypi_0    pypi
[conda] torchvision               0.3.0           py37_cu10.0.130_1    pytorch