[Lot of people have this issue] Porotype still does not load after installing the nightly build

audio
1

image
image945×760 34 KB

Same unsolved issue:

Offical Colab [Same issue and not working]: Google Colab

image
image1122×499 31 KB

Same Github Issue[Unsolved] : ASR Inference tutorial not working · Issue #2336 · pytorch/audio · GitHub

2

You might be missing the import as it works for me:

>>> import torchaudio.prototype
>>> torchaudio.prototype
<module 'torchaudio.prototype' from '/opt/conda/lib/python3.8/site-packages/torchaudio/prototype/__init__.py'>
3

image
image865×687 29.6 KB

If it is working on your colab, do you mind sharing it with me?

4

The import fails with a missing libcudart.so, so I guess you might have updated torchaudio for the wrong CUDA runtime in Colab.
Check the CUDA runtime version via torch.version.cuda and update to the binaries using the same CUDA runtime.

5

That OSError corresponding to the missing shared object file seems to imply a mismatch in torch and torchaudio versions.

The following works for me in the Colab notebook:

!pip3 uninstall -y torch torchvision torchaudio
!pip3 install --pre torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/nightly/cpu

from torchaudio.prototype.ctc_decoder import download_pretrained_files

Start from a fresh/restarted runtime, and run it at the very top.

6

Thank you. It works.

1 Like