---
format: "aidr-story-markdown/v1"
id: "cba58727ea88e064aba5d3e17d40ecf05b2478ea67f245031589c585e0afc5b3"
canonical_url: "https://aidr.today/cba58727?lang=en"
title: "Show HN: I trained a 125M model to autocomplete piano on-device"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-08-20T12:04:38.000Z"
category: "Models"
topics: ["transformer","inference","on-device"]
source_urls: ["https://simedw.com/2026/08/20/midi-autocomplete/","https://news.ycombinator.com/item?id=49373456"]
summary: "Can a small transformer autocomplete MIDI performances in real time?"
---

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

> [Open the canonical story](<https://aidr.today/cba58727?lang=en>)

**Published:** 2026-08-20T12:04:38.000Z
**Category:** Models
**Topics:** transformer, inference, on\-device

## Summary

Can a small transformer autocomplete MIDI performances in real time?

## Sources

- [Story source](<https://simedw.com/2026/08/20/midi-autocomplete/>)
- [Discussion](<https://news.ycombinator.com/item?id=49373456>)