[BUG] Offline does not work

[IgNoras]

Description

Thank you for the product! I have plenty of ideas for implementation, but only in an offline version. Unfortunately, when the internet is disconnected, the container tries to reach "https://registry-1.docker.io/v2/" and won't start anymore. How can I fix this issue?
Thanks in advance!

Environment

• OS: Ubuntu 24, Debian 12
• Browser: Firefox
• Hosting: Local

P.S.: I'm willing to donate for a solution to this problem.

[speedy-rb]

For your specific issue you just need to modify the docker-compose.yml file.

Change pull_policy: always to pull_policy: missing under the whisper configuration.

This assumes you already have the container images on your machine.

But if you want to run whishper offline there are a few more things you need to do.

First you need to make some changes to your docker-compose.yml (on top of changing pull_policy to missing).

1. Change the volumes under the translate image from

- ./whishper_data/libretranslate/data:/home/libretranslate/.local/share

to

- ./whishper_data/libretranslate/data:/root/.local/share

and

- ./whishper_data/libretranslate/cache:/home/libretranslate/.local/cache

to

- ./whishper_data/libretranslate/cache:/root/.local/cache

2. Add a volume under the whishper volume

- ./whishper_data/uploads:/app/models

Then you need to launch the app while you're connected to the internet.

From here do a transcription with a random audio/video with whatever models you want to use (I used tiny and small).

Now you can stop the app (sudo docker compose down) and disconnect from the internet.

For some reason, the whisper transcription models aren't saved to a snapshots folder automatically so you need to do that manually.

For each transcription model do the following:

On your host machine:

cd whishper_data/models

then enter whichever directory you want to work on first

You should be inside a directory like faster-whisper-small.

From here you need to create the snapshot directory

sudo mkdir cache/models--Systran--faster-whisper-<size>/snapshots

Then you need to grab the hash of the model

cat cache/models--Systran--faster-whisper-<size>/refs/main

Take that hash and make a folder with that hash as its name under snapshots

ex.

sudo mkdir cache/models--Systran--faster-whisper-<size>/snapshots/a654c76b4a76c476a54c7b64ca76b4a764cb7

Then you need to copy all the model files into the snapshots folder

sudo cp config.json model.bin tokenizer.json vocabulary.txt cache/models--Systran--faster-whisper-<size>/snapshots/a654c76b4a76c476a54c7b64ca76b4a764cb7/

Repeat this for all the models you want to use offline.

From here you should be able to go offline and use this app (at least I was able to).

It would be great if there was a way to make this app write to snapshots automatically.

It would also be nice to have a docker-compose.yml file that is configured for 100% offline use on this repo.