Apple Releases MLX Framework for Generative AI on GitHub
Amid the raging storm of generative AI, Apple has made a significant move by releasing its MLX framework on GitHub. This open-source array framework is designed for building machine learning transformer language models and text generation AI on Apple’s own silicon. Let’s dive into the details of what this framework entails and how it positions Apple in the competitive AI landscape.
What is Apple’s MLX Framework?
The MLX framework offers a comprehensive set of tools for developers who are building AI models. This includes transformer language model training, large-scale text generation, text fine-tuning, generating images, and speech recognition on Apple silicon. The announcement of the MLX machine learning framework was made by Apple’s machine learning research scientist Awni Hannun on X (formerly Twitter) on Dec. 5.
The MLX framework makes use of various technologies for its different functionalities. It utilizes Meta’s LlaMA and low-rank adoption for text generation, Stability AI’s Stable Diffusion for image generation, and OpenAI’s Whisper for speech recognition. The framework is intended to be familiar to deep learning researchers and is inspired by technologies such as NumPy, PyTorch, Jax, and ArrayFire.
Ease of Use and Efficiency
MLX aims to provide a user-friendly experience while maintaining efficiency in training and deploying models. It is designed to keep arrays in shared memory, allowing supported devices – currently CPUs and GPUs – to run MLX on-device without creating data copies. The Python AI in MLX is said to be familiar to developers who already know how to use NumPy. They can also utilize MLX through a C++ API that mirrors the Python API. The framework also integrates other APIs similar to those used in PyTorch to simplify the process of building complex machine learning models.
Key Features of MLX
The MLX framework comes equipped with composable function transformations, enabling automatic differentiation, vectorization, and computation graph optimization. Computation in MLX is lazy, meaning arrays only materialize when needed, and Apple claims that computation graphing and debugging are “simple and intuitive.” The ultimate goal for MLX, as stated by the Apple developers, is to make it easy for researchers to extend and improve the framework and quickly explore new ideas.
Response from the Industry
NVIDIA AI research scientist Jim Fan praised the release, commending the framework for designing an API familiar to the deep learning audience and showcasing minimalistic examples of OSS models such as Llama, LoRA, Stable Diffusion, and Whisper.
Apple’s Position in the AI Landscape
While Apple has had its artificial intelligence assistant Siri for years, the company seems to be more focused on providing the tools to develop large language models rather than producing the models themselves and the chatbots built with them. However, reports suggest that Apple has been caught off guard by the sudden AI fever in the industry and has been working to catch up. Bloomberg’s Mark Gurman mentioned that Apple is working on upcoming generative AI features for iOS and Siri.
In comparison, Google has recently released its powerful Gemini large language model on the Pixel 8 Pro and in the Bard conversational AI. However, Google still lags behind its rival OpenAI in terms of widespread generative AI functionality.
Conclusion
Apple’s release of the MLX framework represents a significant step in the company’s efforts to tap into the generative AI space. By providing a user-friendly and efficient framework, Apple is poised to empower developers in creating advanced machine learning models. It will be interesting to see how MLX shapes the future of AI development on Apple’s platform.