torchaudio does not work

[mthrok]

Describe the current behavior
The following code was working earlier with PyTorch 1.9,

!pip install torchaudio
import torchaudio

but after #2372, it no longer works with the following error;

Collecting torchaudio
  Downloading torchaudio-0.10.0-cp37-cp37m-manylinux1_x86_64.whl (2.9 MB)
     |████████████████████████████████| 2.9 MB 9.3 MB/s 
Requirement already satisfied: torch==1.10.0 in /usr/local/lib/python3.7/dist-packages (from torchaudio) (1.10.0+cu111)
Requirement already satisfied: typing-extensions in /usr/local/lib/python3.7/dist-packages (from torch==1.10.0->torchaudio) (3.10.0.2)
Installing collected packages: torchaudio
Successfully installed torchaudio-0.10.0
---------------------------------------------------------------------------
OSError                                   Traceback (most recent call last)
<ipython-input-1-90af107c3418> in <module>()
      4 get_ipython().system('pip install torchaudio')
      5 
----> 6 import torchaudio
      7 print(torchaudio.__version__)

4 frames
/usr/lib/python3.7/ctypes/__init__.py in __init__(self, name, mode, handle, use_errno, use_last_error)
    362 
    363         if handle is None:
--> 364             self._handle = _dlopen(self._name, mode)
    365         else:
    366             self._handle = handle

OSError: libcudart.so.10.2: cannot open shared object file: No such file or directory

Describe the expected behavior

a version of torchaudio 0.10 that matches with the pre-installed PyTorch should be installed and works.

What web browser you are using

I tried with Chrome and Safari

• Safari (Version 15.1 (17612.3.2.1.1))
• Chrome (Version 95.0.4638.69 (Official Build) (arm64))

[malfet]

This is sort of expected, as you are trying to mix-n-match torch+cu113 and torchaudio+cu102
Testing if pip install torchaudio -f https://download.pytorch.org/whl/cu113/torch_stable.html will work

And indeed it does:

[mthrok]

Yeah, but the concern is that all the Colab notebooks with torchaudio from the past has stopped working.

[mravanelli]

I have the same issue and this is breaking all the tutorials of SpeechBrain (https://speechbrain.github.io/). It is also breaking the torchaudio tutorials and many others I think. It would be great to fix it such that people can simply do "pip install torchaudio" as before.

[colaboratory-team]

Being worked on with internal bug: b/206623828

[aheba]

Hello @colaboratory-team,
we do a tutorial (20 students) this morning with torchaudio on collab and we got stuck too...

One of the solution is to use:

!pip install torch==1.10.0+cu113 torchvision==0.11.1+cu113 torchaudio==0.10.0+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html 

are you working on that problem ?

[zpapakipos]

Hi, this issue has also broken 50% of our example colabs for the AugLy library (https://github.com/facebookresearch/AugLy/tree/main/examples). Do you have an ETA on when this issue will be fixed, or should we implement workarounds?

[mravanelli]

Right, for us it is a kind of an emergency and many users are complaining about it. I agree It could be important if you can share with us an estimate on when the issue will be fixed.

…

On Thu, 18 Nov 2021 at 08:36, Zoe Papakipos ***@***.***> wrote: Hi, this issue has also broken 50% of our example colabs for the AugLy library (https://github.com/facebookresearch/AugLy/tree/main/examples). Do you have an ETA on when this issue will be fixed, or should we implement workarounds? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2417 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEA2ZVS36SZNZATKKA4ZIHLUMT6PVANCNFSM5H3ZALFA> .

[pierrebedu]

aheba gave a (temporary) solution 23 hours ago

[mravanelli]

Right, but this is quite annoying on our side because we have to basically change something like 30 tutorials (and then modify them again when the issue will be fixed).

…

On Thu, 18 Nov 2021 at 09:42, Pierre Bedu ***@***.***> wrote: aheba gave a (temporary) solution 23 hours ago — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2417 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEA2ZVW6JG2BDADFWXABFITUMUGERANCNFSM5H3ZALFA> .

[colaboratory-team]

Fixed.

[mravanelli]

Thank you very much!

…

On Fri, 19 Nov 2021 at 13:18, colaboratory-team ***@***.***> wrote: Closed #2417 <#2417>. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2417 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEA2ZVVKY3OAWEIB4EFAQN3UM2IGTANCNFSM5H3ZALFA> .

[mthrok]

Thank you @colaboratory-team !

[loretoparisi]

This is amazing!
In my case I had to do the following:

%pip install torch==1.10.0+cu113 torchaudio==0.10.0+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html

Then it worked

import torchaudio
print(torchaudio.__version__)
0.10.0+cu113
time: 699 ms (started: 2021-11-15 12:51:46 +00:00)