This repository contains recipes for training generative music models on top of the Descript Audio Codec.

you can try vampnet in a co-creative looper called unloop. see this link: https://github.com/hugofloresgarcia/unloop

Requires Python 3.9.

you'll need a Python 3.9 environment to run VampNet. This is due to a known issue with madmom.

(for example, using conda)

conda create -n vampnet python=3.9
conda activate vampnet

install VampNet

git clone https://github.com/hugofloresgarcia/vampnet.git
pip install -e ./vampnet

This repository relies on argbind to manage CLIs and config files. Config files are stored in the conf/ folder.

The weights for the models are licensed CC BY-NC-SA 4.0. Likewise, any VampNet models fine-tuned on the pretrained models are also licensed CC BY-NC-SA 4.0.

Download the pretrained models from this link. Then, extract the models to the models/ folder.

You can launch a gradio UI to play with vampnet.

python app.py --args.load conf/interface.yml --Interface.device cuda

To train a model, run the following script:

python scripts/exp/train.py --args.load conf/vampnet.yml --save_path /path/to/checkpoints

for multi-gpu training, use torchrun:

torchrun --nproc_per_node gpu scripts/exp/train.py --args.load conf/vampnet.yml --save_path path/to/ckpt

You can edit conf/vampnet.yml to change the dataset paths or any training hyperparameters.

For coarse2fine models, you can use conf/c2f.yml as a starting configuration.

See python scripts/exp/train.py -h for a list of options.

To debug training, it's easier to debug with 1 gpu and 0 workers

CUDA_VISIBLE_DEVICES=0 python -m pdb scripts/exp/train.py --args.load conf/vampnet.yml --save_path /path/to/checkpoints --num_workers 0

To fine-tune a model, use the script in scripts/exp/fine_tune.py to generate 3 configuration files: c2f.yml, coarse.yml, and interface.yml. The first two are used to fine-tune the coarse and fine models, respectively. The last one is used to launch the gradio interface.

python scripts/exp/fine_tune.py "/path/to/audio1.mp3 /path/to/audio2/ /path/to/audio3.wav" <fine_tune_name>

This will create a folder under conf/<fine_tune_name>/ with the 3 configuration files.

The save_paths will be set to runs/<fine_tune_name>/coarse and runs/<fine_tune_name>/c2f.

launch the coarse job:

python scripts/exp/train.py --args.load conf/generated/<fine_tune_name>/coarse.yml 

this will save the coarse model to runs/<fine_tune_name>/coarse/ckpt/best/.

launch the c2f job:

python  scripts/exp/train.py --args.load conf/generated/<fine_tune_name>/c2f.yml 

launch the interface:

python  app.py --args.load conf/generated/<fine_tune_name>/interface.yml