Learn how to use Apple’s incredible machine-learning framework, MLX, with local models to build a powerful on-device search engine for your text and images.

https://github.com/preternatural-explore/omt-conf-2025

Prerequisites

• Xcode 16.4
• MacOS 15.0 or above (MacOS Sequoia)

Technologies

MLX Swift (GitHub)

• Swift API built on MLX, a high-performance numerical computing framework optimized for Apple silicon.
  • Similar conceptually to libraries like NumPy or PyTorch, but specifically tailored for Swift and Apple's hardware.

MLX Swift Examples (GitHub)

• Complementary repository containing practical implementations built with MLX Swift.
• Demonstrates real-world applications:
  • Embedding models.
  • Large Language Models (LLMs).
  • Vision-Language Models (VLMs).
  • Stable Diffusion and other popular ML architectures.
• Serves as a ready-to-use reference to integrate MLX Swift into actual Swift projects

MLXEmbedders (part of mlx-swift-examples)

• Provides implementations of popular embedding models pre-ported into Swift.
• Allows straightforward integration of embedding models into your Swift projects.
• Facilitates text-to-embedding conversions:
  • Load pre-trained embedding models (e.g., Nomic’s text embeddings).
  • Tokenize text inputs with automatic padding for consistent length.
  • Generate normalized embedding vectors useful for similarity search and semantic matching tasks.
• Simplifies embedding workflows by offering standardized methods and reusable components.