mlx-examples

MLX Examples is a repository of standalone example projects demonstrating the MLX framework, an efficient and flexible machine learning library designed for Apple silicon. The collection spans a wide range of deep learning tasks and model architectures, making it a valuable learning resource for developers and researchers working with MLX. The examples are organized into several categories: **Text Models**: Includes transformer language model training, large-scale text generation with LLaMA and Mistral, mixture-of-experts models (Mixtral 8x7B), parameter-efficient fine-tuning with LoRA/QLoRA, text-to-text multi-task Transformers (T5), and bidirectional language understanding (BERT). **Image Models**: Covers image generation with FLUX and Stable Diffusion/SDXL, image classification using ResNets on CIFAR-10, and convolutional variational autoencoders (CVAE) on MNIST. **Video Models**: Features text-to-video and image-to-video generation with Wan2.1. **Audio Models**: Includes speech recognition with OpenAI's Whisper, audio compression and generation with Meta's EnCodec, and music generation with Meta's MusicGen. **Multimodal Models**: Demonstrates joint text and image embeddings with CLIP, text generation from image and text inputs with LLaVA, and image segmentation with Segment Anything (SAM). **Other Models**: Covers semi-supervised learning on graph-structured data with GCN and normalizing flows for density estimation and sampling. The MNIST example is recommended as a starting point for learning MLX. For more fully featured LLM capabilities, the repository points to MLX LM, a dedicated Python package. Converted model checkpoints are available through the MLX Community organization on Hugging Face, and the project welcomes community contributions of new models. The MLX software suite was initially developed by Awni Hannun, Jagrit Digani, Angelos Katharopoulos, and Ronan Collobert.