How to stop Whisper downloading the model each time

[Dec-35]

I have to run whisper every single day and it downloads the model (medium in my case so pretty long download on slow internet) almost every time I run it. How can I stop that behaviour?

It also gets stuck on 19kiB/s whearas my internet is about 6 or 7 mB/s

[jongwook]

It should not download the model each time once it is fully downloaded. It appears that the Azure CDN that we're using is not particularly friendly with your network. Does it download any faster if you try this address in your browser:

https://openaipublic.azureedge.net/main/whisper/models/345ae4da62f9b3d59415adc60127b97c714f32e89e936602e85993674d08dcb1/medium.pt

Alternatively, you could try this non-CDN address which is typically slower:

https://openaipublic.blob.core.windows.net/main/whisper/models/345ae4da62f9b3d59415adc60127b97c714f32e89e936602e85993674d08dcb1/medium.pt

Once downloaded, Whisper can use the model if you copy them to ~/.cache/whisper/medium.pt.

[lazarok09]

@phineas-pta where is saved in the disk the particular model ?

[vincedani]

👋 I would like to follow the starting guidelines to use the project, but model loading stuck without any output:

$ pip install -U openai-whisper
$ python
>>> import whisper
>>> model = whisper.load_model("tiny.en") % never returns

The load_model somehow uses 100% of my CPU, and changing the device to cuda didn't help me move on. I've also tried to download the model and replace the tiny.en to the specific path, however, load_model blocks the execution for hours.

Do you have any ideas to solve the issue?

[skywalk-explore]

do you have a link the to the small model that can be used on Mac ? The medium model is a bit large and takes too long to process files.

[glangford]

do you have a link the to the small model that can be used on Mac ?

Links to all the models can be found here in the whisper source:

whisper/whisper/__init__.py

Line 17 in ba3f3cd

_MODELS = {

[skywalk-explore]

Thanks man.