how to run the pipeline on the cpu ?

[amitli1]

I'm trying to run pyannote and getting error: "CUDA out of memory".

so, I want to run my code on CPU rather than GPU.

from pyannote.audio import Pipeline

pipeline = Pipeline.from_pretrained("pyannote/speaker-diarization",use_auth_token="token...")

diarization = pipeline("example.wav")

How can I run the speaker-diarization pipeline on the CPU given the above configuration?

[MustafaCQN]

Hi @amitli1

please check this thread
#1095

you can specify the device type as pipeline.to(device)

[amitli1]

It doesn't work:

pipeline = Pipeline.from_pretrained("pyannote/speaker-diarization",use_auth_token=MY_TOKEN)
pipeline.to("cpu")

Error:
AttributeError: 'SpeakerDiarization' object has no attribute 'to'

I'm using the following versions:

pyannote.audio==2.1.1
pyannote.core==4.5
pyannote.database==4.1.3
pyannote.metrics==3.2.1
pyannote.pipeline==2.3

[MustafaCQN]

Can you share your
print(torch.__version__)
and
print(torch.cuda.is_available()
sorry for earlier answer, it will work on offlline models. Seems you are using an online model

[amitli1]

Results fot the versions & cuda:

1.13.1+cu117
True

How can I use pyannote pipeline offlline models ?

[MustafaCQN]

How can I use pyannote pipeline offlline models ?

You can use the Segmentation model. Downloadable from HuggingFace.

Results fot the versions & cuda:

1.13.1+cu117
True

You can do a tweak in this operation. You can delete your torch and re-install the cpu only version in order to run it from cpu

[amitli1]

Segmentation model is just one model of the pyannote/speaker-diarization (which contains Segmentation + Embedding + Clusterring) .
So download just it from HuggingFace is not sufficient, am I right ?