Show HN: I trained a 125M model to autocomplete piano on-device

Community post

TL;DR: I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The biggest improvements came from finding the right MIDI representation, cleaning the training data aggressively, and adding DPO post-training.

WebNovare Imports W
Show HN: I trained a 125M model to autocomplete piano on-device

TL;DR: I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The biggest improvements came from finding the right MIDI representation, cleaning the training data aggressively, and adding DPO post-training.

Source: simedw.com (529 points, 110 comments) Hacker News discussion

Reader Support

Support WebNovare

Send a crypto tip if this article helped you.

Comments

No comments yet.

Log in or register to join the conversation.