TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Jun Kim, creator and maintainer of the oMLX MLX-based inference engine, has joined Hugging Face. Hugging Face says oMLX will gain stability and faster development as a fully funded project, while remaining Apache 2.0 under Kim’s leadership.
Jun Kim, the creator and maintainer of the open-source MLX inference engine oMLX, has joined Hugging Face to work on Apple’s MLX ecosystem, the company announced on its blog. The move turns oMLX from a side project into a fully maintained and funded effort, with Hugging Face positioning the hire as part of its broader investment in local, on-device AI.
Hugging Face described the hire as a continuation of its long-standing support for MLX, Apple’s array-based machine learning framework optimized for Apple Silicon. The company said it has backed the framework since its release in late 2023 — a release it referred to as “the Christmas present from Awni and Angelos in 2023” — and noted that Hugging Face’s Hub is where users find and contribute MLX models.
For oMLX itself, Hugging Face said the expected impact is stability and faster development. The company stated that graduating oMLX “from a side job to a fully maintained and funded project” will let Kim guide contributors more effectively and build for the long term. Two things will not change, according to the announcement: oMLX remains Apache 2.0 licensed, and Kim continues to lead the project as before.
Hugging Face also outlined a concrete technical focus: streamlining the transition from a transformers model definition to a reference MLX implementation that different inference engines can consume. The company said the transformers library has become the reference for machine learning model definitions, and the goal is to make it easier for new transformers models to run on MLX.
Jun Kim, oMLX Creator, Joins Hugging Face to Support the MLX Community
The maintainer of the open-source MLX inference engine oMLX moves from side project to a fully maintained and funded effort — staying Apache 2.0, under Kim’s continued leadership, as Hugging Face doubles down on local, on-device AI.
We are completely invested in local AI, and MLX is a central piece of the ecosystem.
— Hugging Face, announcement blog postWhat Changes — and What Doesn’t
Hugging Face positions the hire as a continuation of its support for MLX since the framework’s release in late 2023. Two things are explicitly guaranteed to stay the same; several things get stronger.
Apache 2.0 License
oMLX remains Apache 2.0 licensed. Hugging Face stated this directly in the announcement — no relicensing, no strings attached to the code.
Kim Keeps Leading
Jun Kim continues to lead the project as before, now better positioned to guide contributors and build for the long term with institutional backing.
Stability & Speed
Graduating “from a side job to a fully maintained and funded project” means greater stability and faster development for oMLX users and contributors.
MLX, oMLX and the Hub
MLX is Apple’s array-based machine learning framework optimized for Apple Silicon. Around it sits a layer cake of community projects — with Hugging Face’s Hub as the distribution point where users discover and publish MLX-format models.
| Project | Role in the stack | Relationship to Hugging Face |
|---|---|---|
| MLX | Apple’s ML framework for Apple Silicon, released end of 2023 by Apple engineers. | Supported since release — “the Christmas present from Awni and Angelos in 2023.” |
| oMLX | Open-source inference engine created and maintained by Jun Kim; a testbed for new ideas. | Now fully funded and maintained — its creator joins the company. |
| mlx-lm | Foundational library providing language model implementations on top of MLX. | Existing collaboration; Hugging Face hopes to strengthen ties and upstream work where it makes sense. |
| mlx-vlm | Foundational library for vision-language model implementations on MLX. | Named as a target for closer collaboration with Cheng, Prince, Yagil and their teams. |
| LMStudio | Local model runtime consuming MLX-format models. | Cited among MLX-related projects Hugging Face has collaborated with. |
| The Hub | Distribution point where users find and contribute MLX models. | Hugging Face’s own platform; MLX model support will continue as local AI grows. |
From transformers to MLX, Faster
The one specific technical commitment: streamlining the path from a transformers model definition to a reference MLX implementation that multiple inference engines can consume — making it easier for new transformers models to run on MLX.
transformers definition
The transformers library has become the reference for machine learning model definitions.
Streamlined conversion
A quick, repeatable path from model definition to a reference MLX implementation.
Reference MLX code
A canonical implementation multiple inference engines can consume — not just oMLX.
Local inference
Unblocking the community to run local AI in any shape or form on Apple Silicon.
“Graduating from a side job to a fully maintained and funded project will allow Jun to better guide the contributors and build for the long-term.”
— Hugging Face blog post“Our end goal is to unblock the community to run local AI in any shape or form, and provide the tools and building blocks to make that happen.”
— Hugging Face blog postWhat Would Confirm the Promise
- Release cadence: watch the oMLX repository and release notes for the promised stability work and faster shipping now that maintenance is funded.
- Conversion path: a key milestone is a public, reusable transformers-to-MLX conversion path usable by multiple engines.
- Shared infrastructure: whether collaboration with mlx-lm, mlx-vlm, LMStudio and the named MLX teams produces common tooling.
- Upstreaming: whether oMLX work flows back into the foundational libraries it depends on, or stays siloed around the Hub.
The Essentials, Answered
What is oMLX?
An open-source inference engine built for Apple’s MLX framework, optimized for Apple Silicon. Created and maintained by Jun Kim, who now works on it full time at Hugging Face.
Does the license change?
No. Hugging Face stated directly that oMLX stays Apache 2.0 and that Jun Kim continues to lead the project as before.
What is MLX?
Apple’s machine learning framework, released at the end of 2023, designed to run efficiently on Apple Silicon for local model training and inference.
What changes for users?
Greater stability and faster development as oMLX moves from a personal side effort to a fully maintained and funded project. Specific feature timelines were not announced.
Why Local AI on Apple Silicon Gains
The hire signals that Hugging Face sees local, on-device AI as a growth area rather than a niche. According to the company, usage of open, local AI is accelerating, and MLX is a central piece of that ecosystem on Apple hardware. Funding a dedicated maintainer for an MLX inference engine gives developers working on Apple Silicon a better-supported path to run models without cloud dependencies.
For the oMLX community, the practical benefit is continuity with more resources. Maintainer burnout and abandonment are persistent risks for single-maintainer open-source projects; moving the work into a funded role addresses that risk directly. Hugging Frame also framed oMLX as a testbed for new ideas that builds on foundational libraries such as mlx-lm and mlx-vlm, with a stated willingness to upstream work “to wherever it makes sense” — meaning improvements could flow into the wider MLX stack rather than staying siloed.
Top picks for "omlx creator maintainer"
As an affiliate, we earn on qualifying purchases.
MLX, oMLX and the Hub
MLX is Apple’s machine learning framework designed for Apple Silicon, released at the end of 2023 by Apple engineers. It has since become a common foundation for running and training models locally on Macs and other Apple devices. Community projects such as mlx-lm and mlx-vlm provide model implementations on top of the framework.
oMLX is an open-source inference engine in that ecosystem, created and maintained by Jun Kim. Hugging Face’s Hub serves as a distribution point where users discover MLX-format models and publish their own. Hugging Face said it has collaborated with MLX-related projects including mlx-lm, mlx-vlm, and LMStudio, and hopes to strengthen relationships with the teams behind those projects — naming Cheng, Prince, Yagil and their teams — to serve the community jointly.
“We are completely invested in local AI, and MLX is a central piece of the ecosystem.”
— Hugging Face, announcement blog post
Open Questions About the Arrangement
Several details remain unspecified in the announcement. Hugging Face did not state Jun Kim’s exact job title, team placement, or start date, nor did it describe the financial terms or scope of the funding behind oMLX’s new status.
The company also did not provide a timeline for its stated technical goals — such as the streamlined transformers-to-MLX conversion path — or name specific upstream contributions that are planned. The announcement expresses intent (“we’d love to upstream work to wherever it makes sense”) rather than committed deliverables, so the pace of visible change to oMLX and the wider MLX stack is not yet clear.
Roadmap Signs to Watch
Watch the oMLX repository and release notes for signs of the promised stability work and faster release cadence now that the project has funded maintenance. A key milestone to track is progress on the stated goal of a quick path from transformers model definitions to reference MLX implementations usable by multiple engines.
Further signals will include whether Hugging Face’s collaboration with mlx-lm, mlx-vlm, LMStudio and the named MLX teams produces shared infrastructure, and whether work from oMLX is upstreamed into the foundational libraries it depends on. Hugging Face said it will continue supporting MLX models on the Hub as local AI usage grows.
Where I land
I read this as a modest but genuinely useful piece of open-source news: a single-maintainer project gets funded institutional backing, and the company involved has committed — in writing — to preserving the license and the maintainer’s leadership. That combination is rarer than it should be, and it addresses the most common failure mode for community infrastructure projects.
The strongest counterargument is that corporate sponsorship of a previously independent project can shift its priorities toward the sponsor’s platform goals — in this case, Hugging Face’s Hub and transformers-centric workflow — at the expense of the broader MLX community. The announcement’s emphasis on upstreaming work and collaborating with other MLX teams is a stated intent, not a delivered outcome.
What would change my assessment: if oMLX releases slow rather than accelerate in the coming quarters, or if contributions remain Hub-centric without flowing back into mlx-lm and mlx-vlm, that would suggest the funding is consolidating rather than strengthening the ecosystem. Conversely, a public, reusable transformers-to-MLX conversion path would confirm the announcement’s most concrete promise.
Source: Hugging Face
Key Questions
What is oMLX?
oMLX is an open-source inference engine built for Apple’s MLX framework, which is optimized for Apple Silicon. It was created and has been maintained by Jun Kim, who now joins Hugging Face to work on it full time.
Does oMLX’s license change now that its maintainer joined Hugging Face?
No. Hugging Face stated directly that oMLX stays Apache 2.0 and that Jun Kim continues to lead the project as before.
What is MLX?
MLX is Apple’s machine learning framework, released at the end of 2023, designed to run efficiently on Apple Silicon. It is widely used for running and training AI models locally on Apple hardware.
What changes for oMLX users?
According to Hugging Face, users should see greater stability and faster development, as the project moves from a personal side effort to a fully maintained and funded one. Specific feature timelines were not announced.
Why is Hugging Face investing in MLX?
The company said it sees local, on-device AI as an accelerating area and views MLX as a central part of that ecosystem on Apple hardware. Hugging Face’s Hub also hosts the MLX models the community publishes and downloads.
Source: Hugging Face
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
