# ARahim3/mlx-tune

Fine-tune LLMs on your Mac with Apple Silicon. SFT, DPO, GRPO, Vision, TTS, STT, Embedding, and OCR fine-tuning — natively on MLX. Unsloth-compatible API.

Repository: https://github.com/ARahim3/mlx-tune
Canonical: https://ross.abutalabs.com/products/mlx-tune
Homepage: https://arahim3.github.io/mlx-tune/
Language: Python
License: Apache-2.0
License Family: permissive
Topics: apple-silicon, deep-learning, huggingface, large-language-models, llm, llm-finetuning, local-llm, lora, machine-learning, macos, mlx, on-device-ai, peft, transformers, unsloth, vision-language-model, speech-recognition, speech-to-text, text-to-speech, whisper
Last push: 2026-06-23T12:24:30+00:00

## Health v2 (maintenance only)
Score: 75/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 89, release rhythm 90, longevity 17
- inputs: {"age_days": 242, "days_push": 71, "days_rel": 71, "gap_med": 1, "n_releases_24m": 34}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1389, forks 91 (observed 2026-08-28T04:04:35.323788+00:00)

## What it is
A Python library for fine-tuning LLMs, vision-language, audio (TTS/STT), embedding, OCR, and JEPA models natively on Apple Silicon Macs using Apple's MLX framework. It provides an Unsloth-compatible API so existing Unsloth training scripts run on Macs with only an import change, and exports to HuggingFace or GGUF formats.

## Use cases
- fine-tune an LLM with LoRA on my MacBook without a cloud GPU
- run my existing Unsloth training script on Apple Silicon
- fine-tune Whisper for speech-to-text locally on a Mac
- train a TTS model with LoRA on Apple Silicon
- fine-tune a vision-language model like Qwen on my Mac
- train sentence embeddings for semantic search on-device
- fine-tune an OCR model for receipts or handwriting recognition
- prototype fine-tuning locally then move the same script to a CUDA cluster

## When to choose
- you want to fine-tune LLMs or multimodal models locally on an M1-M5 Mac without cloud GPUs
- you already use Unsloth and want code portability between Mac prototyping and CUDA training
- you need LoRA fine-tuning across many modalities (text, vision, audio, OCR, embeddings) in one library
- you want to export fine-tuned models to GGUF for Ollama or llama.cpp

## When to avoid
- you need maximum training throughput on large-scale clusters — CUDA-based tools like Unsloth are the gold standard
- you are on Linux or Windows with NVIDIA GPUs — this is Apple Silicon only
- you need full fine-tuning of very large models beyond unified memory limits
- you require a battle-tested, officially supported tool — this is an unofficial community project

## Facets
- artifact type: library
- maturity: active
- function: llm-training, machine-learning, deep-learning, speech-recognition, tts, ocr, transformers
- domain: large-language-models, machine-learning, deep-learning, speech-processing, computer-vision
- platform: python
- tags: mlx, apple-silicon, lora, peft, fine-tuning, unsloth-compatible, sft, dpo, grpo, vision-language-model, whisper, gguf-export, huggingface, on-device-ai, jepa, mixture-of-experts, natural-language-processing, macos, gpu

## Member repositories
- ARahim3/mlx-tune (main) score 75

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:35.323788+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:39:40.912410+00:00, confidence not recorded.
  - readme: https://github.com/ARahim3/mlx-tune (fetched 2026-08-28T04:04:35.323788+00:00, sha 6df1b0cd6e1f)
  - homepage: https://arahim3.github.io/mlx-tune/ (fetched 2026-08-29T11:54:54.785596+00:00, sha d17e58e0ae99)
  - registry_pypi: https://pypi.org/pypi/mlx-tune/json (fetched 2026-08-29T11:54:54.795223+00:00, sha d8acc95deb2f)
- Data as of 2026-08-30T08:39:29.467469+00:00.