How to load model from disk ?

[amitli1]

Im loading the silero-vad model using the following code:

   vad, vad_utils = torch.hub.load(repo_or_dir='snakers4/silero-vad',
                                    model='silero_vad',
                                    force_reload=False,
                                    onnx=False)

Once I did it, is it possible to load the model and it's utils from disk (and not from torch.hub) ?
How can I do it ?

[snakers4]

Hi,

Two simplest ways:

• Load the model via hub once, then use the cache flag in torch.hub.load;

• Clone the repo somewhere, then pass this path as a parameter to torch.hu.load, there should be a parameter like cache path or similar;

[snakers4]

Here - https://pytorch.org/docs/stable/hub.html#torch.hub.load

[mukundt]

@snakers4 does torch.hub.load load the model into CPU memory, or does it keep using the file on disk? I'm noticing that the VAD runs very slowly on a machine with an HDD, compared to an SSD.

[snakers4]

Torch hub loads model into RAM. Vad speed should not depend on disk speed, unless you write/ read from disk extensively.