Emily - A new MLX-based backend for Nx

ausimian

Emily is an Elixir library that runs Nx computations on Apple’s MLX. Install it as the default Nx backend and Nx, defn, Axon, Nx.Serving, and Bumblebee all dispatch to the Metal GPU on Apple Silicon with no further integration work.

Highlights:

Nx.Backend with transparent fallback for uncovered ops
Nx.Defn.Compiler that threads Bumblebee and Nx.Serving through MLX
Fused transformer kernels (rms_norm, layer_norm, rope, scaled-dot-product attention), optionally applied to any Bumblebee model via a shim
Affine group-wise int2 / int4 / int8 quantisation
bf16 mixed-precision training
Per-process Metal command queues for concurrent inference on a shared model
Runnable Livebooks for DistilBERT QA, Qwen3 quantized generation, MNIST training, Whisper transcription,
and the fused kernels

MLX is vendored in (for now) because the recent command queue threading-related features have not yet been released. The first compile will compile MLX, which can be a bit slow (> 1 min).

macOS / Apple Silicon. Requires Xcode with the Metal toolchain;

https://github.com/ausimian/emily

19 comments

#library #libraries #emily

26 2194 19

2026-07-28 07:39:04 UTC

Showing Posts 1 to 10

Show Best Posts
Show All Posts (oldest first)
Show All Posts (newest first)

kip

ex_cldr Core Team

Amazing work, that’s such a great contribution and I can’t wait to give it a spin.

Post #1

LostKobrakai

How does this compare to GitHub - elixir-nx/emlx: MLX Backend for Nx · GitHub?

Post #2

ausimian

Roughly:

emily exposes the concurrency model now available in mlx, I assume emlx can’t do this because of the version it uses.
emily has features I don’t believe emlx advertises. Fused kernels, quantisation, mixed precision training (might be wrong about this one), maybe also zero-copy to_binary/1.
emily vendors mlx at a specific commit (for reasons outlined above and in more details in the docs), emlx uses a relatively old binary
There’s quite a few conformance tests in emily that use real open-source models, but I mostly did that for my own sanity

Post #3

borodark

Can’t wait to get my hands on Apple M hardware to integrate this with my eXMC.

Post #4

ausimian

Also Emily allows you to control whether MLX’s kernels are compiled AOT or JIT. The latter significantly reduces the size of the final application, at some small trade-off for start-up time. The CI tests, including conformance, run in both modes.

Post #5

christhekeele

Nice! Does this mean that there’s a backend for Nx that now Just Works™️ in Livebooks run on OSX that is actually GPU accellerated?

Post #6

ausimian

Yes, there a few livebooks in the docs that demonstrate usage - it’s really just setting the default backend in the Mix.install config e.g:

Post #7

polvalente

Nx Core Team

We’re open to bringing in any features to EMLX you feel are missing (and also you as a maintainer) if you wanna join efforts!

Post #8

polvalente

Nx Core Team

I will say it would have been nice to acknowledge EMLX in the README as a source of inspiration, since the PLAN.md committed to main references it quite a bit.

Post #9