MLX Installation Guide

Prerequisites

Set up Xcode properly on your Apple Silicon Mac before installing MLX.

1

Install Xcode

Open the App Store
Search for "Xcode" and install
Wait for download completion (several GB)

2

Accept License

Open Xcode once to accept the initial license agreements.

3

Configure Command Line Tools

$ sudo xcode-select --switch /Applications/Xcode.app/Contents/Developer Password: ******** $ sudo xcodebuild -license accept License accepted

Build from Source

Once Xcode is configured, build MLX from source.

$ rm -rf build && mkdir build && cd build $ cmake .. -- The C compiler identification is AppleClang 15.0.0.15000040 -- The CXX compiler identification is AppleClang 15.0.0.15000040 -- Found Metal: /System/Library/Frameworks/Metal.framework -- Configuring done -- Build files have been written to: /Users/username/mlx/build $ make -j8 [ 1%] Building CXX object CMakeFiles/mlx.dir/mlx/backend/metal/kernels.cpp.o ... [100%] Built target metal_capture

Verify Installation

Test the build to ensure everything works correctly.

$ ./tests/tests Running all tests. .................... All tests passed!

$ ./examples/cpp/tutorial array([[1, 1], [1, 1]], dtype=float32)

Build Artifacts

The build process creates several important components:

Core Library

libmlx.a - Main MLX static library

Example Programs

linear_regression - Linear regression example
logistic_regression - Logistic regression example
tutorial - Basic operations tutorial
metal_capture - Metal performance profiling
distributed - Distributed computing example

Test Suite

tests/tests - Comprehensive test executable

System Configuration

Build SettingsCMAKE_BUILD_TYPE=Release
MLX_BUILD_TESTS=ON
MLX_BUILD_EXAMPLES=ON
MLX_BUILD_METAL=ON
MLX_BUILD_CPU=ON

System Requirements

macOS 13.5 or later (Sonoma recommended)
Python 3.9 or later
Apple Silicon (M1/M2/M3/M4)
8GB RAM minimum (16GB+ recommended)

Key Features

Unified Memory: Zero-copy CPU/GPU data sharing
Lazy Evaluation: Compute-on-demand optimization
Metal Integration: Native Apple GPU support
Multi-Device: Seamless CPU/GPU execution

Usage Examples

Basic Python Usage

$ python3 -c "import mlx.core as mx; print(mx.zeros((2,2)))" array([[0., 0.], [0., 0.]])

Performance Profiling

$ MTL_CAPTURE_ENABLED=1 python3 your_script.py Capturing Metal trace for analysis

Environment Variables

DEVICE=gpu|cpu - Select compute device
MTL_CAPTURE_ENABLED=1 - Enable Metal capture
LOW_MEMORY=1 - Optimize for low memory systems

Testing Framework

Run All Tests

$ python scripts/run_tests.py 🚀 Starting MLX Test Suite ✅ All required dependencies found 🔍 Running C++ tests... 🐍 Running Python tests... All tests completed successfully

Test Categories

CPU Tests: Basic array operations and math functions
GPU Tests: Metal-accelerated computations
Distributed Tests: Multi-node operations
Integration Tests: End-to-end workflows