Hasty Briefsbeta

Bilingual

Show HN: We made our own inference engine for Apple Silicon

10 months ago
  • #AI
  • #Apple Silicon
  • #Inference Engine
  • High-performance inference engine for AI models on Apple Silicon.
  • Simple, high-level API with hybrid architecture (GPU kernels or MPSGraph).
  • Unified model configurations for easy addition of new models.
  • Traceable computations for correctness against source-of-truth implementation.
  • Utilizes unified memory on Apple devices.
  • Add uzu dependency via Cargo.toml.
  • Create an inference Session with model and configuration.
  • Supports CLI mode with commands like run and serve.
  • Uses its own model format; export models using lalamo.
  • Prebuilt Swift framework (uzu-swift) available for SPM.
  • Licensed under MIT License.