MIDIT — Checkpoints

Pretrained weights for MIDIT, a BERT-style masked language model for symbolic music infilling (REMI+ tokenization, ~42.5M parameters).

Code, usage and full results: https://github.com/CSCPadova/MIDIT
Online demo: https://huggingface.co/spaces/SimoneZanetti00/MIDIT-demo
Thesis (PDF): https://huggingface.co/SimoneZanetti00/MIDIT-checkpoints/blob/main/MIDIT_thesis.pdf
Presentation video (audio examples): https://www.youtube.com/watch?v=ywxU9QwpMJg

File	Phase	Description
`midit-phase1-epoch14.ckpt`	Phase 1 (MLM pre-training)	base / zero-shot model, val_loss 2.7254
`midit-infilling-epoch05.ckpt`	Phase 1b (BarFillMasking fine-tuning)	fine-tuned infilling model

Weights contain the Lightning state_dict + hyper_parameters only (optimizer state removed). Load with the model class from the repository.

Trained on the Lakh MIDI Dataset; released for research purposes.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

SimoneZanetti00
/

MIDIT-checkpoints