# MLX-VLM

A package for running inference and fine-tuning Vision Language Models, including omni models with audio and video support, on Apple Silicon Macs using MLX.

MLX-VLM supports many vision-language and OCR-focused models, including DeepSeek-OCR, DOTS-OCR, ERNIE 4.5 VL, GLM-OCR, Phi-4, MiniCPM, PaddleOCR-VL, Gemma 4 and LLaVA-OneVision, with a command-line interface for inference, thinking-budget control, speculative decoding, a Gradio chat UI, a FastAPI server, continuous batching, automatic prefix caching, KV cache quantization and multi-image chat.

It ships an agent-skills bundle under `skills/` for common MLX-VLM workflows, covering usage, conversion, development and support, which loads into coding agents such as Claude Code, Codex or Gemini so the agent follows the project's conventions instead of guessing.

It installs with `pip install mlx-vlm`, with an optional `[ui]` extra for the Gradio interface; it is written in Python and licensed under MIT.

Website: https://github.com/Blaizzy/mlx-vlm