mlx-engine
The Apple MLX inference engine bundled with LM Studio, supporting text and vision model inference and speculative decoding on Apple Silicon.
mlx-engine is the MLX-based inference engine used by LM Studio to run large language and vision-language models on Apple Silicon Macs. It is built on the mlx-lm inference engine, the Outlines library for structured LLM output, and mlx-vlm for vision-model inference, and ships pre-bundled with LM Studio 0.3.4 and newer.
Outside of LM Studio, it can be run standalone with a demo script that loads a model with the lms CLI and runs text generation, vision-model prompts over one or more images (supporting models such as Llama-3.2-Vision, Pixtral, Qwen2-VL, and Llava-v1.6), or speculative decoding. It requires macOS 14 (Sonoma) or later and Python 3.11.
It is released under the MIT license.