sander-wood/tunesformer · Apply for community grant: Academic project

Dear Huggingface team,

I am writing to apply for a community GPU grant for TunesFormer, a project I am currently working on as a PhD student. As an avid user of the Huggingface platform, I have found it to be an incredibly useful resource for my research and personal projects. I am excited to continue showcasing my work, especially in the field of symbolic music, on this platform.

TunesFormer is a project that allows users to generate symbolic music by inputting control codes to set the musical form. It is based on the GPT2-small architecture and trained on the Massive ABC Notation Dataset. While the model is capable of generating high-quality music, the current CPU resources available for the project are limited and result in very slow generation speeds (up to 10 minutes). As a student with limited financial resources, it is difficult for me to justify paying for Huggingface's GPU resources.

Therefore, I am hoping to receive a community GPU grant from Huggingface to allow me to provide a faster and more convenient interface for users to generate symbolic music using TunesFormer. The grant would not be used for training the model, but solely for providing online interfaces for users to generate music. I believe that this project has the potential to benefit a wide range of users who are interested in exploring the possibilities of symbolic music generation.

Thank you for your time and consideration.

Sincerely,
Shangda Wu