We can fine-tune VampNet models using LoRA (with an R=8), by fine-tuning on small audio datasets ranging in size from a couple of minutes to a couple of hours. Here are some unconditional samples generated by fine-tuning VampNet on some small datasets of music:

A model trained on a breakbeats / step playlist:

https://open.spotify.com/playlist/64o9tPBetUfXC449OKnbGI?si=ATL35_LxS3OQ11SujWAaKA

Unconditional Samples

6.wav

5.wav

4 copy.wav

7.wav

3 copy.wav

2.wav

Trio Los Panchos

https://open.spotify.com/playlist/37i9dQZF1DZ06evO2bgbPo?si=4484fbf0aded4a7f

Unconditional Samples

3.wav

2.wav

1.wav