spellbrush/PupuM2D

Frequency-Aware Self-Supervised Music Representation Learning

This is the official Hugging Face model repository for the paper "Frequency-Aware Self-Supervised Music Representation Learning".

• Project Page: https://www.yichenggu.com/PupuM2D/
• GitHub Repository: https://github.com/sizigi/PupuM2D/

Model Checkpoints

This repository contains the following checkpoints:

Model Name	Directory	Description
PupuM2D-Tiny	./PupuM2DV2_25Hz_tiny/*	5M parameter tiny version of PupuM2D.
PupuM2D-Small	./PupuM2DV2_25Hz_small/*	22M parameter small version of PupuM2D.
PupuM2D-Base	./PupuM2DV2_25Hz_base/*	86M parameter base version of PupuM2D.
PupuM2D-Large	./PupuM2DV2_25Hz_large/*	307M parameter large version of PupuM2D.
PupuM2D-Huge	./PupuM2DV2_25Hz_huge/*	632M parameter huge version of PupuM2D.
PupuM2D-giant	./PupuM2DV2_25Hz_giant/*	1.1B parameter giant version of PupuM2D.

How to use

You need to put the pretrained models in:

  PupuM2D/experiments

of our official repository.

Citation

@article{PupuM2D,
  title        = {Frequency-Aware Self-Supervised Music Representation Learning},
  author       = {Yicheng Gu and Junan Zhang and Jerry Li and Zhizheng Wu and Lauri Juvela},
  year         = {2026},
  journal      = {arXiv:2606.25713},
}