训练了一个125M参数模型，在端侧实时自动补全钢琴MIDI
有人训练了一个125M参数的模型，能在iPhone上实时自动补全钢琴MIDI音符，并做成了配套App。 个人项目，涉及端侧小模型推理和音乐生成交叉领域。125M参数量在移动端实时运行是个有意思的工程选择，适合对边缘AI和音乐技术交叉感兴趣的读者。
RSS: https://simedw.com/feed.xml
Personal blog by Simon Edwardsson - dd if=/dev/me of=/simedw.com
提交点赞
评论
点赞
收藏How to force AI agents to use an egress proxy
How we make a proxy the only way out of an AI agent sandbox.
Building a Magic Eye Generator and Decoder
Magic Eye autostereograms encode depth using repeated pixel patterns. This post shows how to generate them and how to recover the depth map.
I trained a 9M speech model to fix my Mandarin tones
Built this because tones are killing my spoken Mandarin and I can't reliably hear my own 's a 9M Conformer-CTC model trained on ~300h (AISHELL + Primewords), quantized to INT8 (11 MB), runs 100% in-br...
Coding Agents Are Good First-Time User Testers
Coding agents encode years of software conventions. Their confusion is often a signal that your interface is asking too much of a new user.
Learning a Language Using Only Words You Know
A proof-of-concept language learning app that uses LLMs to generate definitions of unknown words using only previously mastered vocabulary.
You Need an Ideation Channel
Product decisions shouldn't be democratic, but great ideas can come from anywhere.
Is Gemini 2.5 good at bounding boxes?
I decided to write a small benchmark and check Gemini 2.5 performance on MS-COCO, focusing on object detection.
Spegel, a Terminal Browser That Uses LLMs to Rewrite Webpages
Spegel is a proof-of-concept terminal web browser that feeds HTML through an LLM before serving it as markdown in the terminal.
Books are malleable
What if we could modify books as easily as we tweak open-source software? Perhaps we already can.