Code for Yeerchiu/MIDI-TEXT
Adding text input for MIDI transformer model
MIDI-Text pair data generation utilize of MU-LLaMA
if you want to run app.py, at least one ckpt file is needed above(Yeerchiu/MIDI-TEXT).
- At present, the code is run in an environment with GPU. There is no proof yet to see if it can run with only CPU.
- The average GPU memory usage during the runtime of this model is about 5-6GB
- Due to lack of computing resources and time, only 20,000+ MIDI files collected from Free MIDI Web were used on the MIDI-Text pair training.
- There may be some address errors that have not been corrected.
- The text can influence the model generation but cannot always get the desirable results, optimization is still needed.
- Follow the official guidance for configuring the GPU graphics drivers, CUDA, and cuDNN, etc.
- Create a new environment
conda create midi -n python=3.10 conda activate midi
- Install Pytorch following the https://pytorch.org/
- Install packages
conda install --yes --file requirements.txt
or
pip install -r requirements.txt
Original Model
Training Dataset
MIDI-Text pair data generated on V100
Training on RTX 4090
App Tested in RTX 4060 (Laptop)