Skip to main content

​
Installation

MLX is available on PyPI with platform-specific packages for macOS and Linux. Choose the installation method that matches your platform and requirements.

​
macOS (Apple Silicon)

MLX runs natively on Apple silicon Macs using the Metal GPU backend.

​
Requirements

  • M-series chip (Apple silicon)
  • macOS >= 14.0
  • Python >= 3.10 (native, not x86 via Rosetta)

​
Install with pip

pip install mlx
MLX is only available on devices running macOS 14.0 or higher.

​
Verify Python Architecture

Ensure you’re using native Python (not running via Rosetta): 
python -c "import platform; print(platform.processor())"
The output should be arm. If it shows i386, you’re using a non-native Python. Switch to a native Python installation using Conda or another package manager.

​
Linux

MLX supports Linux with CUDA GPU acceleration or CPU-only execution.

​
CUDA Backend

For NVIDIA GPUs with CUDA support:

​
Requirements

  • NVIDIA architecture >= SM 7.5
  • NVIDIA driver >= 550.54.14
  • CUDA toolkit >= 12.0
  • Linux distribution with glibc >= 2.35
  • Python >= 3.10

​
Install with pip

pip install mlx[cuda12]
CUDA 13 requires an NVIDIA driver >= 580 or an appropriate CUDA compatibility package.

​
CPU-Only Backend

For running MLX on Linux without GPU acceleration:

​
Requirements

  • Linux distribution with glibc >= 2.35
  • Python >= 3.10

​
Install with pip

pip install mlx[cpu]

​
Troubleshooting

If your OS and Python versions meet the requirements but pip still fails, you’re likely using a non-native Python.Check your Python architecture:
python -c "import platform; print(platform.processor())"
On macOS with Apple silicon, this should output arm. If it shows i386, switch to a native Python installation.
Ensure you’re using Python 3.10 or later:
python --version
On macOS, verify you’re running macOS 14.0 or higher:
sw_vers

​
Building from Source

For advanced users who need to build from source, MLX supports custom builds with various configuration options.

​
Requirements

  • C++ compiler with C++20 support (e.g., Clang >= 15.0)
  • CMake >= 3.25
  • macOS: Xcode >= 15.0 with macOS SDK >= 14.0
  • Linux (CPU): BLAS and LAPACK libraries
  • Linux (CUDA): CUDA toolkit and cuDNN

​
Clone and Build

1

Clone the repository

git clone [email protected]:ml-explore/mlx.git mlx && cd mlx
2

Install with pip

For development:
pip install -e ".[dev]"
For CUDA on Linux:
CMAKE_ARGS="-DMLX_BUILD_CUDA=ON" pip install -e ".[dev]"
3

Run tests

python -m unittest discover python/tests
For detailed C++ build instructions and advanced configuration options, see the Build from Source documentation.

​
Next Steps

Quick Start Tutorial

Learn how to create your first MLX arrays and perform basic operations

Build docs developers (and LLMs) love

Get started for freeTalk to us