[seamless_m4t] Skip some tests when speech is not available #38430
Conversation
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
|
cc @ydshieh |
| """ | ||
| Decorator marking a test that requires speech. These tests are skipped when speech isn't available. | ||
| """ | ||
| return unittest.skipUnless(is_speech_available(), "test requires speech")(test_case) |
There was a problem hiding this comment.
Here we have to decide what we really require.
def is_speech_available():
# For now this depends on torchaudio but the exact dependency might evolve in the future.
return _torchaudio_available
and in setup.py we have
extras["speech"] = deps_list("torchaudio") + extras["audio"]
If we want to require torchaudio + audio for the tests, then is_speech_available is not enough.
Also the message test requires speech will be confusing as speech is not a python library.
If we only want to require torchaudio, then we have def require_torchaudio(test_case):.
cc @eustlb
There was a problem hiding this comment.
Yes, it would be nice to figure out if is_speech_available is needed or redundant. If it isn't, I will change the decorator to require_torchaudio. I have no idea on this though.
As for the message, we can change it to when the packages required for speech models are not available ?
There was a problem hiding this comment.
I am happy to merge for now if the message is changed to
test requires torchaudio
And come back to check the details later.
This PR adds the
require_speechdecorator and adds it to three tests inseamless_m4tandseamless_m4t_v2that fail with an ImportError ifis_speech_available() == False