unsloth-mlx 0.3.5

MLX-powered LLM fine-tuning for Apple Silicon - A drop-in replacement for Unsloth

Unsloth-MLX

Fine-tune LLMs on your Mac with Apple Silicon
Prototype locally, scale to cloud. Same code, just change the import.

Quick Start · Training Methods · Examples · Status

---

[!NOTE] Why I Built This (A Personal Note)

I rely on Unsloth for my daily fine-tuning on cloud GPUs—it's the gold standard for me. But recently, I started working on a MacBook M4 and hit a friction point: I wanted to prototype locally on my Mac, then scale up to the cloud without rewriting my entire training script.

Since Unsloth relies on Triton (which Macs don't have, yet), I couldn't use it locally. I built unsloth-mlx to solve this specific "Context Switch" problem. It wraps Apple's native MLX framework in an Unsloth-compatible API.

The goal isn't to replace Unsloth or claim superior performance. The goal is code portability: allowing you to write FastLanguageModel code once on your Mac, test it, and then push that exact same script to a CUDA cluster. It solves a workflow problem, not just a hardware one.

This is an "unofficial" project built by a fan, for fans who happen to use Macs. It's helping me personally, and if it helps others like me, then I'll have my satisfaction.

Why Unsloth-MLX?

Bringing the Unsloth experience to Mac users via Apple's MLX framework.

• 🚀 Fine-tune LLMs locally on your Mac (M1/M2/M3/M4/M5)
• 💾 Leverage unified memory (up to 512GB on Mac Studio)
• 🔄 Same API as Unsloth - your existing code just works!
• 📦 Export anywhere - HuggingFace format, GGUF for Ollama/llama.cpp

# Unsloth (CUDA)                        # Unsloth-MLX (Apple Silicon)
from unsloth import FastLanguageModel   from unsloth_mlx import FastLanguageModel
from trl import SFTTrainer              from unsloth_mlx import SFTTrainer

# Rest of your code stays exactly the same!

What This Is (and Isn't)

This is NOT a replacement for Unsloth or an attempt to compete with it. Unsloth is incredible - it's the gold standard for efficient LLM fine-tuning on CUDA.

This IS a bridge for Mac users who want to:

• 🧪 Prototype locally - Experiment with fine-tuning before committing to cloud GPU costs
• 📚 Learn & iterate - Develop your training pipeline with fast local feedback loops
• 🔄 Then scale up - Move to cloud NVIDIA GPUs + original Unsloth for production training

Local Mac (Unsloth-MLX)     →     Cloud GPU (Unsloth)
   Prototype & experiment          Full-scale training
   Small datasets                  Large datasets
   Quick iterations                Production runs

Project Status

🚀 v0.3.5 - Merged model save + load_adapter fixed!

Feature	Status	Notes
SFT Training	✅ Stable	Native MLX training
Model Loading	✅ Stable	Any HuggingFace model (quantized & non-quantized)
Save/Export	✅ Stable	HF format, GGUF (see limitations)
DPO Training	✅ Stable	Full DPO loss
ORPO Training	✅ Stable	Full ORPO loss
GRPO Training	✅ Stable	Multi-generation + reward
KTO/SimPO	✅ Stable	Proper loss implementations
Chat Templates	✅ Stable	15 models (llama, gemma, qwen, phi, mistral)
Response-Only Training	✅ Stable	train_on_responses_only()
Multi-turn Merging	✅ NEW	to_sharegpt() + conversation_extension
Column Mapping	✅ NEW	apply_column_mapping() auto-rename
Dataset Config	✅ NEW	HFDatasetConfig structured loading
Vision Models	⚠️ Beta	Via mlx-vlm
PyPI Package	✅ Available	uv pip install unsloth-mlx

Installation

# Using uv (recommended - faster and more reliable)
uv pip install unsloth-mlx

# Or using pip
pip install unsloth-mlx

# From source (for development)
git clone https://github.com/ARahim3/unsloth-mlx.git
cd unsloth-mlx
uv pip install -e .

Quick Start

from unsloth_mlx import FastLanguageModel, SFTTrainer, SFTConfig
from datasets import load_dataset

# Load any HuggingFace model (1B model for quick start)
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="mlx-community/Llama-3.2-1B-Instruct-4bit",
    max_seq_length=2048,
    load_in_4bit=True,
)

# Add LoRA adapters
model = FastLanguageModel.get_peft_model(
    model,
    r=16,
    target_modules=["q_proj", "k_proj", "v_proj", "o_proj"],
    lora_alpha=16,
)

# Load a dataset (or create your own)
dataset = load_dataset("yahma/alpaca-cleaned", split="train[:100]")

# Train with SFTTrainer (same API as TRL!)
trainer = SFTTrainer(
    model=model,
    train_dataset=dataset,
    tokenizer=tokenizer,
    args=SFTConfig(
        output_dir="outputs",
        per_device_train_batch_size=2,
        learning_rate=2e-4,
        max_steps=50,
    ),
)
trainer.train()

# Save (same API as Unsloth!)
model.save_pretrained("lora_model")  # Adapters only
model.save_pretrained_merged("merged", tokenizer)  # Full model
model.save_pretrained_gguf("model", tokenizer)  # GGUF (see note below)

[!NOTE] GGUF Export: Works with non-quantized base models. If using a 4-bit model (like above), see Known Limitations for workarounds.

Chat Templates & Response-Only Training

from unsloth_mlx import get_chat_template, train_on_responses_only

# Apply chat template (supports llama-3, gemma, qwen, phi, mistral, etc.)
tokenizer = get_chat_template(tokenizer, chat_template="llama-3")

# Or auto-detect from model name
tokenizer = get_chat_template(tokenizer, chat_template="auto")

# Train only on responses (not prompts) - more efficient!
trainer = train_on_responses_only(
    trainer,
    instruction_part="<|start_header_id|>user<|end_header_id|>\n\n",
    response_part="<|start_header_id|>assistant<|end_header_id|>\n\n",
)

Supported Training Methods

Method	Trainer	Implementation	Use Case
SFT	SFTTrainer	✅ Native MLX	Instruction fine-tuning
DPO	DPOTrainer	✅ Native MLX	Preference learning (proper log-prob loss)
ORPO	ORPOTrainer	✅ Native MLX	Combined SFT + odds ratio preference
GRPO	GRPOTrainer	✅ Native MLX	Reasoning with multi-generation (DeepSeek R1 style)
KTO	KTOTrainer	✅ Native MLX	Kahneman-Tversky optimization
SimPO	SimPOTrainer	✅ Native MLX	Simple preference optimization
VLM	VLMSFTTrainer	⚠️ Beta	Vision-Language models

Examples

Check examples/ for working code:

• Basic model loading and inference
• Complete SFT fine-tuning pipeline
• RL training methods (DPO, GRPO, ORPO)

Requirements

• Hardware: Apple Silicon Mac (M1/M2/M3/M4/M5)
• OS: macOS 13.0+ (15.0+ recommended for large models)
• Memory: 16GB+ unified RAM (32GB+ for 7B+ models)
• Python: 3.9+

Comparison with Unsloth

Feature	Unsloth (CUDA)	Unsloth-MLX
Platform	NVIDIA GPUs	Apple Silicon
Backend	Triton Kernels	MLX Framework
Memory	VRAM (limited)	Unified (up to 512GB)
API	Original	100% Compatible
Best For	Production training	Local dev, large models

Known Limitations

GGUF Export from Quantized Models

The Issue: GGUF export (save_pretrained_gguf) doesn't work directly with quantized (4-bit) base models. This is a known limitation in mlx-lm, not unsloth-mlx.

What Works:

• ✅ Training with quantized models (QLoRA) - works perfectly
• ✅ Saving adapters (save_pretrained) - works
• ✅ Saving merged model (save_pretrained_merged) - works
• ✅ Inference with trained model - works
• ❌ GGUF export from quantized base model - mlx-lm limitation

Workarounds:

1. Use a non-quantized base model (recommended for GGUF export):

   # Use fp16 model instead of 4-bit
   model, tokenizer = FastLanguageModel.from_pretrained(
       model_name="mlx-community/Llama-3.2-1B-Instruct",  # NOT -4bit
       max_seq_length=2048,
       load_in_4bit=False,  # Train in fp16
   )
   # Train normally, then export
   model.save_pretrained_gguf("model", tokenizer)  # Works!
   

2. Dequantize during export (results in large fp16 file):

   model.save_pretrained_gguf("model", tokenizer, dequantize=True)
   # Then re-quantize with llama.cpp:
   # ./llama-quantize model.gguf model-q4_k_m.gguf Q4_K_M
   

3. Skip GGUF, use MLX format: If you only need the model for MLX/Python inference, just use save_pretrained_merged() - no GGUF needed.

Related Issues:

• mlx-lm #353 - MLX to GGUF conversion
• mlx-examples #1382 - Quantized to GGUF

Contributing

Contributions welcome! Areas that need help:

• Custom MLX kernels for even faster training
• More comprehensive test coverage
• Documentation and examples
• Testing on different M-series chips (M1, M2, M3, M4, M5)
• VLM training improvements

License

Apache 2.0 - See LICENSE file.

Acknowledgments

• Unsloth - The original, incredible CUDA library
• MLX - Apple's ML framework
• MLX-LM - LLM utilities for MLX
• MLX-VLM - Vision model support

---

Community project, not affiliated with Unsloth AI or Apple.
⭐ Star this repo if you find it useful!