The MLX framework, a set of tools for building machine learning transformer language models and text generation AI, has been released by Apple on GitHub. This open source array framework is designed to work on Apple’s own silicon and provides developers with the ability to build AI models, perform large-scale text generation, fine-tune text, generate images, and enable speech recognition. MLX utilizes various technologies such as Meta’s LlaMA for text generation, low-rank adoption for text generation, Stability AI’s Stable Diffusion for image generation, and OpenAI’s Whisper for speech recognition.
MLX takes inspiration from NumPy, PyTorch, Jax, and ArrayFire, but distinguishes itself by keeping arrays in shared memory, allowing for efficient on-device execution without creating data copies. Apple aims to make MLX accessible to developers familiar with NumPy, offering a Python AI that can also be used through a C++ API. The MLX framework simplifies the building of complex machine learning models by providing APIs similar to those used in PyTorch, along with built-in function transformations for differentiation, vectorization, and computation graph optimization. It is designed to be user-friendly while maintaining efficiency in training and deploying models.
NVIDIA AI research scientist Jim Fan praised Apple’s design, stating that MLX provides a familiar API for the deep learning audience and showcases examples of OSS models that are widely popular. Apple’s focus seems to be on providing tools for building large language models, rather than developing the models themselves. However, Bloomberg’s Mark Gurman reported that Apple executives have been working on catching up with the AI trend, indicating that Apple is also developing generative AI features for iOS and Siri. In comparison, Google is still behind OpenAI in terms of widespread generative AI functionality, even though it recently released its powerful Gemini large language model.
Overall, Apple’s release of the MLX framework on GitHub demonstrates the company’s commitment to supporting developers in the machine learning space. By providing a user-friendly yet efficient framework, Apple aims to encourage researchers to explore new ideas and quickly develop and deploy machine learning models.
(Note: The word count of this article is 362 words)