Create a pop music Transformer

:wave: Please read the topic category description to understand what this is all about


If you treat music notation as a form of “text”, you can use language modelling to generate new songs! The goal of this project is to explore how well Transformers perform at music modelling.


None that we could find on the Hub, but see here for some pretrained music Transformers


None that we could find on the Hugging Face Hub, but see here for some ideas


This task probably involves pretraining a Transformer, which can potentially take multiple days using the GPU resources provided by AWS. An interesting alternative would be to see whether one can integrate an existing pretrained model within :hugs: Transformers and use that to either fine-tune.

Desired project outcomes

  • Create a Streamlit or Gradio app on :hugs: Spaces that people can remix famous songs with newly generated ones
  • Don’t forget to push all your models and datasets to the Hub so others can build on them!

Additional resources

Discord channel

To chat and organise with other people interested in this project, head over to our Discord and:

  • Follow the instructions on the #join-course channel

  • Join the #pop-music-transformer channel

Just make sure you comment here to indicate that you’ll be contributing to this project :slight_smile:

1 Like

This one seems pretty difficult, but too interesting to pass up. I’ll try working on it.


Hey @psistolar ! Yes this is a somewhat challenging project and one that might not be possible in the timeframe if you adopt the pretraining approach. Let me think about how we can simplify it and I’ll post back here :slight_smile:

In the meantime, there’s a new Discord channel (see topic description) in case you and others need it