mlx-lm mit Homebrew installieren

Projektlinks

• Homepage

Befehle und Aliase

• mlx_lm
• mlx_lm.awq
• mlx_lm.benchmark
• mlx_lm.cache_prompt
• mlx_lm.chat
• mlx_lm.convert
• mlx_lm.dwq
• mlx_lm.dynamic_quant
• mlx_lm.evaluate
• mlx_lm.fuse
• mlx_lm.generate
• mlx_lm.gptq
• mlx_lm.lora
• mlx_lm.manage
• mlx_lm.perplexity
• mlx_lm.server
• mlx_lm.share
• mlx_lm.upload

Projektgeschichte

Its ecosystem role is to make MLX usable for Hugging Face-style LLM work without each user wiring model loading, tokenization, generation loops, quantization, and fine-tuning scripts by hand. The README highlights Hugging Face Hub integration, quantizing and uploading models, low-rank and full fine-tuning with quantized models, and distributed inference and fine-tuning through MLX distributed support. The MLX examples repository points users from earlier LLM examples to MLX LM as the more fully featured package.

Wie es verwendet wird

Adoption is tied to the Apple-silicon local-model community. The MLX Community organization on Hugging Face hosts ready-to-use model weights for MLX LM and related MLX projects, and Apple's WWDC material presents MLX LM as the package for running and fine-tuning large language models on Macs. In Homebrew it is a convenience package for users who want the MLX LLM command suite installed as normal CLI tools rather than only through Python packaging.

Quellen

• https://developer.apple.com/videos/play/wwdc2025/298/
• https://formulae.brew.sh/formula/mlx-lm
• https://github.com/ml-explore/mlx-examples/blob/main/llms/README.md
• https://github.com/ml-explore/mlx-lm
• https://huggingface.co/mlx-community