Apple just discreetly launched a raft of new machine learning tools, and they’re free

Apple has launched a series of open source tools as part of a move to provide machine learning capabilities on Apple Silicon. 

Known as Apple MLX, the machine learning framework has been developed by the tech giant’s ML research division and will enable users to build AI tools harnessing its in-house silicon.

The framework is available through open source libraries such as PyPI and GitHub, Apple confirmed.

“MLX is designed by machine learning researchers for machine learning researchers,” Apple said on GitHub. “The framework is intended to be user-friendly, but still efficient to train and deploy models.”

“The design of the framework itself is also conceptually simple. We intend to make it easy for researchers to extend and improve MLX with the goal of quickly exploring new ideas.”

Apple said the design of MLX is “inspired by frameworks like PyTorch, Jax, and ArrayFile”, yet boasts notable differences. Key among these is the use of its unified memory model, the firm said, with arrays hosted in shared memory.

This means operations on MLX arrays can be “performed on any of the support device types without performing data copies”.

The MLX examples repo, accessible via GitHub, enables a range of capabilities, including the ability to conduct language model training, large-scale text generation using Meta’s Llama model, and speech recognition via OpenAI’s whisper model.

Users can also generate images with Stable Diffusion, the firm noted.

In a post on X, Apple ML researcher Awni Hannun said MLX is designed specifically for use on Apple Silicon, such as MacBooks, pointing toward the company’s potential goal of broadening access to tools by harnessing its own silicon.

Apple MLX: Key features for users

Key features of Apple MLX include the ability to leverage familiar APIs and dynamic graph construction, as well as multi-device capabilities. 

“MLX has a Python API that closely follows NumPy,” the firm’s GitHub repo explains. “MLX also has a fully featured C++ API, which closely mirrors the Python API.”

Dynamic graph construction has also been simplified, Apple said, providing developers with more optimized capabilities to streamline debugging.

“Computation graphs in MLX are built dynamically. Changing the shapes of function arguments does not trigger slow compilations, and debugging is simple and intuitive.”

Apple’s discreet AI movements

The roll-out of Apple MLX coincided with the launch of Google’s flagship Gemini AI model, which appears to have stolen the limelight. 

However, this move does show the tech giant has been quietly innovating on the AI front amid the furor of the generative AI explosion over the last year.

In stark contrast to Google, Microsoft, and other industry heavyweights, Apple has been keeping its cards close to its chest during this period.

Reports emerged in July 2023 that the company was developing its own internal large language model (LLM), dubbed ‘Apple GPT’, but as of yet nothing has been confirmed.

CEO Tim Cook has been relatively subdued on the topic also. However, in an August interview with CNBC said the tech giant does intend to integrate generative AI capabilities across its product offerings.

“We view AI and ML as fundamental core technologies. And they are virtually embedded in every product that we build,” he said.

Repeated speculation over the company’s approach to AI has been further fueled by the fact that Q3 earnings showed the firm increased R&D spending by roughly $3 billion up to that point compared to the same period in 2022.

In August, ITPro reported that Apple has been keen to emphasize the heightened need for locally-run AI models by leveraging its in-house silicon capabilities.