Tutorial to train a GPT-2 model from scratch using Hugingg Face and publishing it as a Gradio demo using Spaces. The model generates guitar music. For encoding the guitar MIDI files of the Mutopia Project, I am using the excellent implementation of Dr. Tristan Beheren of the paper: MMM: Exploring Conditional Multi-Track Music Generation with the Transformer.
By the end of this tutorial, you should have a Gradio demo similar to this one.
To start the tutorial, please visit the first notebook: 1. Collecting the data.
The dataset is built from the Mutopia Project.
This tutorial is a work in progress. You can take a look at Hugging Face at the following:
After generating the music in the Hugging Face widget, you can listen to the results using this notebook.