#1717th of 47 in Model serving
mlx-lm
Python package and CLI for running and fine-tuning LLMs on Apple silicon with MLX
- Stars
- 7.4k
- License
- MIT
- Last commit
- Oct 2026
- Last release
- Oct 2026
- Language
- Python
4,280 on 1 Apr 20267,431 on 21 Oct 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
MLX LM generates text, chats in a terminal REPL, and fine-tunes LLMs on Apple silicon using the MLX framework. It loads models from the Hugging Face Hub, quantizes and uploads them, and supports LoRA and full fine-tuning, including on quantized models. It ships a Python API and command-line tools such as mlx_lm.generate, mlx_lm.chat and mlx_lm.server.
Who it is for: Developers on Macs running or fine-tuning LLMs locally with MLX
Strengths
- Pulls MLX-compatible models from Hugging Face, with thousands in the MLX Community org
- Converts, quantizes (e.g. 4-bit) and uploads models with one command
- LoRA and full fine-tuning, including on quantized models
- Rotating KV cache, prompt caching and configurable prefill step size for long contexts
Weaknesses
- Targets Apple silicon only, per the README
- Wiring memory for large models needs macOS 15 or higher
- Unsupported model architectures need an issue or a pull request
- Some tokenizers require enabling trust_remote_code
What it needs
- Needs MLX, Hugging Face Hub
- Models: Hugging Face Hub models in MLX format, quantized models, LoRA, full fine-tuning
Also in Model serving
See all 47| Rank | Project | Score |
|---|---|---|
| 1 | LocalAISelf-hosted inference server with OpenAI-compatible API and pluggable backends | 83 out of 100 |
| 2 | llama.cppC/C++ inference engine serving GGUF models over an OpenAI-compatible API | 81 out of 100 |
| 3 | OllamaLocal model runner with CLI and REST API for open-weight LLMs | 80 out of 100 |
| #4 | strataRuns a 125B-parameter MoE model on one consumer GPU plus system RAM | 79 out of 100 |
| #5 | vLLMInference engine for Hugging Face LLMs with PagedAttention and continuous batching | 77 out of 100 |