Article directory
Overview
Apple is giving developers new tools for training and running large language models on its hardware.
Mainly introducing
Development concept
Apple is paying homage to the field of open source artificial intelligence by releasing a new series of open source artificial intelligence tools.
The MLX framework for machine learning was developed specifically for Apple's Silicon processor architecture. MLX is available via GitHub and is designed to simplify machine learning model training and deployment on Apple hardware.
Development details
Apple said the design of MLX was "inspired" by other popular frameworks, includingPyTorch, Jax and ArrayFire. However, MLX differs in the unified memory model - arrays in MLX reside in shared memory, and operations can be performed on any supported device type without performing data copies.
The MLX repository reads: “The framework is designed to be user-friendly yet still efficient in training and deploying models. The framework itself is also designed to be conceptually simple. We intend to make it easier for researchers to extend and improve MLX to Explore new ideas quickly."
MLX has a Python API that closely follows NumPy, a popular Python programming library. MLX also contains a C++ API and higher-level packages that follow PyTorch to build more complex models.
MLX features
Other key features of MLX include:
-
Composable function transformations: MLX features composable function transformations for automatic differentiation, automatic vectorization, and computational graph optimization.
-
Lazy calculations: Calculations in MLX are lazy and arrays are only materialized when needed.
-
Dynamic graph construction: Computational graphs in MLX are dynamically constructed. Parameters do not trigger slow compilation if the user changes the shape of the function, making debugging easier and more intuitive.
use
To demonstrate the capabilities of MLX, Apple machine learning research scientist Awni Hannun posted a video on X (Twitter) showing a 7 billion parameter version of Meta's LLaMA running on the M2 Ultra chip, which is used in Apple's high-end computing systems Available in Mac Studio and Mac Pro.
Other examples of Apple demonstrating MLX include using stable diffusion to generate images, using OpenAI’s Whisper for speech recognition, and using LoRA for efficient parameter fine-tuning.
Like most companies, Apple has been ramping up its artificial intelligence efforts. Reports emerged this summer that the company was developing its own web app-based chatbot service **"Apple GPT"**
Can MLX be used in commercial software?
Apple's MLX is provided under the MIT License, which permits broad freedom of use,including commercial use .
It is a condition of the license that a copyright notice and a permission notice be included in all copies of the software.