🛠️ I trained a 125M model to autocomplete piano on-device
Summary
A developer trained a 125M-parameter transformer model to autocomplete piano performances in real time on-device, achieving about 108 notes per second on an iPhone 15. The system acts like GitHub Copilot by taking a few notes played on a MIDI piano as a prompt and continuing the performance locally.
Why it’s interesting
It brings real-time, on-device AI code-completion style functionality to musical performances using a custom-trained transformer model running locally on a smartphone.
Target user
Piano players and musicians using MIDI keyboards
Business model
Free
Source metrics: Points 48 · Comments 9
HN discussion · Project
Source: Hacker News / Show HN
Summary
A developer trained a 125M-parameter transformer model to autocomplete piano performances in real time on-device, achieving about 108 notes per second on an iPhone 15. The system acts like GitHub Copilot by taking a few notes played on a MIDI piano as a prompt and continuing the performance locally.
Why it’s interesting
It brings real-time, on-device AI code-completion style functionality to musical performances using a custom-trained transformer model running locally on a smartphone.
Target user
Piano players and musicians using MIDI keyboards
Business model
Free
Source metrics: Points 48 · Comments 9
HN discussion · Project
Source: Hacker News / Show HN