Machine Learning Library
1.2k 2026-04-18
ARahim3/mlx-tune
Enables efficient fine-tuning of LLMs, Vision, Audio, and OCR models on Apple Silicon Macs with an Unsloth-compatible API.
Core Features
Fine-tune various AI models (LLMs, Vision, Audio, OCR) natively on Apple Silicon Macs.
Offers an Unsloth-compatible API for seamless code portability between Mac and cloud GPUs.
Leverages Apple's MLX framework and unified memory for efficient local prototyping.
Supports multiple fine-tuning methods including SFT, DPO, and GRPO.
Allows export of fine-tuned models to HuggingFace format and GGUF for Ollama/llama.cpp.
Quick Start
pip install mlx-tuneDetailed Introduction
mlx-tune is an unofficial project designed to bridge the gap for Mac users who want to fine-tune large language models and other AI models locally on Apple Silicon. It wraps Apple's native MLX framework with an Unsloth-compatible API, allowing developers to prototype and iterate on their Macs with fast local feedback loops. The primary goal is code portability, enabling users to write training scripts once and seamlessly transition them to CUDA clusters with Unsloth for production-scale training, solving the 'Context Switch' problem without aiming to replace Unsloth's cloud performance.