Model components / Inference code

MLX LM

MLX LM generates text with large language models on Apple silicon using MLX, pulling any of thousands of compatible models from the Hugging Face Hub. It ships CLI commands for generation, chat, a server and quantized model conversion, with prompt caching, a rotating KV cache and distributed inference and fine-tuning through mx.distributed. Apple's ml-explore organization maintains it as the LM layer split out of the core mlx repo.

Verified 2026-09-01 via GitHub, the LICENSE body, the repository README, the PyPI project JSON and pypistats.

Openness

5 high confidence

5.0

license: MIT(OSI)
source: public
core-gated: ungated(no enterprise or paid tier

The LICENSE body is the stock MIT text, "Copyright (c) 2023 Apple Inc.", read in full. The repository is public and unarchived and builds the published package; nothing in the README names a paid tier, hosted build or withheld component, so the core reads as ungated.

https://raw.githubusercontent.com/ml-explore/mlx-lm/main/LICENSE recorded 2026-09-01
LICENSE body, read in full: stock MIT text, "Copyright (c) 2023 Apple Inc.", no appended condition.
https://api.github.com/repos/ml-explore/mlx-lm recorded 2026-09-01
Repo metadata for ml-explore/mlx-lm - archived false, license MIT, 6,853 stars, pushed 2026-09-01.
https://raw.githubusercontent.com/ml-explore/mlx-lm/main/README.md recorded 2026-09-01
README, read for a paid tier or gated build (none found) and for the product's shape: "MLX LM is a Python package for generating text and fine-tuning large language models on Apple silicon with MLX"; generation/chat/server CLI, HF Hub integration, quantization and upload, distributed inference and fine-tuning with mx.distributed.

Adoption

4 high confidence

4.0

PyPI mlx-lm drew 1,124,056 downloads in the trailing 30 days, banding at level 4 (1M-10M). The package backlinks to ml-explore/mlx-lm. conda-forge distribution exists and is uncounted, so the figure is a floor on one channel of two; it does not change the band's direction.

https://pypistats.org/api/packages/mlx-lm/recent recorded 2026-09-01
"last_month":1124056
https://pypi.org/pypi/mlx-lm/json recorded 2026-09-01
home_page = https://github.com/ml-explore/mlx-lm; version 0.31.3; summary "LLMs with MLX and the Hugging Face Hub".

Capability

3 medium confidence

3.0

Two bands below the vllm anchor, level with uzu, llamafile and the other single-target local engines: it serves one hardware family and carries none of the datacenter parallelism or disaggregation surface the 4s and 5s here are placed on. Within the Apple-silicon niche it is the reference LM runtime, which the adoption axis, not this one, credits.

https://raw.githubusercontent.com/ml-explore/mlx-lm/main/README.md recorded 2026-09-01
README feature list: HF Hub integration, quantization and upload, LoRA and full fine-tuning, distributed inference and fine-tuning with mx.distributed, prompt caching, rotating fixed-size KV cache, prefill step size, mlx_lm.server.

Verified 2026-09-01