rave-training

Utilities and experiments for training RAVE.

Currently, A10 instances on Lambda Cloud (cloud.lambdalabs.com) are the target deployment environment.

Setup

Start your A10 instance on Lambda Cloud, creating and downloading a PEM file key if you haven't already
Once running, launch the instance's Cloud IDE (JupyterLab)
Make sure your audio data is in a directory called /home/ubuntu/training-data. You can do this e.g. by using the JupyterLab upload feature.
Create a new terminal in JupyterLab
Clone this repository: git clone https://github.com/becker929/rave-training.git && cd rave-training
Make script executable: chmod +x ./setup-rave-lambdalabs.sh
Run script ./setup-rave-lambdalabs.sh
To begin training, replace <TRAINING-RUN-NAME> and run the command: nohup /home/ubuntu/.pyenv/versions/3.10.11/bin/python3.10 ./rave-training.py --name="<TRAINING-RUN-NAME>" &
Confirm that RAVE is training
1. You may need to press return or open a new terminal
2. Run tail -f /home/ubuntu/rave-training/nohup.out
In a terminal, run tensorboard --logdir runs --port 6080
In your local terminal, run ssh -i ~/path/to/lambda-key.pem -N -f -L localhost:16080:localhost:6080 ubuntu@<your.instance.ip>
Now, you should be able to view the progress of the training via tensorboard by visiting localhost:16080 in your browser
To export for use in MaxMSP etc., run /home/ubuntu/.pyenv/versions/3.10.11/bin/rave export --run="/home/ubuntu/runs/<RUN-FOLDER>" --streaming
1. Other uses may require that you remove the --streaming flag. See the official RAVE Readme for details.

Too complicated? Reach out to me and I may be able to help.

The rave subfolder contains a fork of RAVE/rave.

Local changes to rave:

Todo: