Wrong torchaudio version is installed into docker images

[tstadel]

Describe the bug
When using torchaudio in docker image deepset/haystack:base-gpu-v1.13.1, we get the following error:

Error message

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/opt/venv/lib/python3.10/site-packages/torchaudio/__init__.py", line 1, in <module>
    from torchaudio import (  # noqa: F401
  File "/opt/venv/lib/python3.10/site-packages/torchaudio/_extension.py", line 136, in <module>
    _check_cuda_version()
  File "/opt/venv/lib/python3.10/site-packages/torchaudio/_extension.py", line 128, in _check_cuda_version
    raise RuntimeError(
RuntimeError: Detected that PyTorch and TorchAudio were compiled with different CUDA versions. PyTorch has CUDA version 11.6 whereas TorchAudio has CUDA version 11.7. Please install the TorchAudio version that matches your PyTorch version.

Expected behavior
Image uses the correct torchaudio version.

Additional context
Add any other context about the problem here, like document types / preprocessing steps / settings of reader etc.

To Reproduce

import torchaudio

FAQ Check

• Have you had a look at our new FAQ page?

[tstadel]

I seems to be solved by adding --extra-index-url https://download.pytorch.org/whl/cu116 to the torchaudio install command in the Dockerfile.

[sjrl]

I think this would be better solved by combining these two lines

haystack/docker/Dockerfile.base

Lines 33 to 34 in 5689c43

	pip install --no-cache-dir -U torchaudio && \
	pip install --no-cache-dir .${haystack_extras} && \

in our docker file into one pip install statement. Or change the order, so we only install pytorch audio after pytorch has already been installed through the haystack_extras.

I believe pip handles these dependency conflicts better when everything is installed together.