SBX work meeting autumn 2004

Two hands-on examples for the SBX work meeting

Running a whisper model on wombat

Hint: Don't do this in you home folder! Use e.g. /var/tmp/<your_name> instead. And don't forget to clean up!

1. Clone repository and enter container directory git clone https://github.com/spraakbanken/work_meeting_autumn_2024 && cd work_meeting_autumn_2024/whisper
2. touch .nobackup
3. Create virtual environment virtualenv env && . ./env/bin/activate or python -m venv env && source env/bin/activate
4. Install pytorch pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121 (You might have to use TMP_DIR=/var/tmp pip if you happen to run out of spae in /tmp)
5. Install all other dependencies pip install -r requirements.txt
6. Download demo data bash download.sh
7. Run transcription LD_PRELOAD=./env/lib/python3.9/site-packages/nvidia/cudnn/lib/libcudnn.so.9 python transcribe.py input.mp3

Deploying a container on demo

1. Clone repository and enter container directory git clone https://github.com/spraakbanken/work_meeting_autumn_2024 && cd work_meeting_autumn_2024/container
2. Build image `podman compose build
3. Start container in the background podman compose up -d
4. Exclude from backup touch ~/.local/share/containers/.nobackup
5. Enable linger, i.e. user processes are not automatically ended when logging out, loginctl enable-linger
6. create .htaccess file e.g. in /export/htdocs/herbert/container`

RewriteRule ^(.*)$ http://localhost:5000/$1 [P]
Header edit* Location "http://localhost:5000/" "http://demo.spraakdata.gu.se/herbert/container"