Jun Kim is an open-source ML engineer at Hugging Face and creator of oMLX. The project runs language models on Apple Silicon Macs with model management, request batching and a reusable cache.[1]
Start with omlx →
Links
Jun Kim: GitHub Jun Kim: personal biography 
X Hugging Face Open-Source ML Engineer · Hugging Face
Career
Kim’s personal biography describes earlier global business-planning work at a Korean food company, including building pipelines and cross-market data systems. Hugging Face announced in September 2026 that he had joined to continue leading oMLX and support the MLX ecosystem. The announcement says oMLX remains Apache 2.0 licensed and presents the move as support for maintenance and collaboration, rather than closing the project.[4][3]
Contributions
oMLX combines continuous batching with a two-tier cache for the key/value state created while processing prompts. Frequently reused blocks stay in memory; colder blocks can move to SSD. Reusing that state avoids recomputing the same context for every request. The app also manages several models, allowing everyday models to remain loaded while larger ones are swapped as needed. It exposes an OpenAI-compatible endpoint and a macOS management interface, making local models usable from existing clients. These features concern inference on supported Macs, not training a new foundation model.[2]
Tools, writing & talks
Tools
Start here
omlxgithub.comAn Apple Silicon inference server with continuous batching, tiered prompt caching, model management and compatible client endpoints.
Begin with the oMLX quick start on a supported Apple Silicon Mac and one modest model. Confirm your client can reach the local endpoint before comparing caching or batching behavior on longer sessions.
Creator and lead maintainer[2][3]Original source
Writing, research & talks
- Hugging Face welcomes Jun Kimhuggingface.co
The employer’s account of funding oMLX maintenance and improving collaboration across the MLX ecosystem.
Subject and listed co-author[3]