metal-gpu

Apple Neural Engine (ANE) LLM inference engine — reverse-engineered private APIs, Metal GPU shaders, hybrid ANE+GPU+CPU on Apple Silicon. 32 tok/s matching llama.cpp, 3.6 TFLOPS fused ANE mega-kernels.

The NVIDIA Container Toolkit for Mac — Give any Docker container full Apple Silicon Metal GPU access. 100+ GPU operations, LLM inference, training, image gen, audio, embeddings. Zero CUDA. Just Metal.

Run AI models too large for your Mac's memory — at near-full speed. Intelligent expert caching, speculative execution, and 15+ research techniques for MoE inference on Apple Silicon.

Open quantum simulation network for Apple Silicon

This repository replicates the NemoStation/Marlin-2B paradigm natively on Apple Silicon using the MLX framework, achieving ultra-fast temporal video captioning and sub-second visual grounding entirely within Apple's Unified Memory space.

GPU-accelerated quantum circuit simulator for Apple Silicon (MLX) with Google Willow, IBM Heron, and QuTech Tuna-9 noise models. Up to 34 qubits on M2 Ultra.

Production RAG system for scientific literature synthesis with SPECTER2 embeddings, Metal GPU acceleration, multi-LLM support, and automatic BibTeX citations.

GPU-accelerated robot motion planning on Apple Silicon. Port of NVIDIA cuRobo (CUDA) to MLX — real-time collision-free trajectory generation on M-series Macs.

High-performance C++ simulation framework for modeling active Brownian particles under cylindrical confinement in 3D space. Features dual implementations with standard OpenMP parallelization for cross-platform compatibility and Metal GPU acceleration for Apple Silicon, achieving up to 27× speedup for large-scale systems (N≥1000 particles).

A high-performance C++ simulation framework for modeling biolocomotion and premelting dynamics in ice using finite difference methods. This implementation numerically solves coupled partial differential equations for chemotaxis and diffusion: via OpenMP SIMD and via Metal for Apple Silicon.

Free, self-hosted, ultra-low-latency Push-to-Talk voice typing & AI translation for Windows, offloaded to an Apple Silicon Mac via Apple MLX & Ollama.

Sovereign, 100% on-device, offline-first AI Image Generation Studio engineered natively for macOS on Apple Silicon (M1/M2/M3/M4) via Apple Metal GPU & MLX architecture. Powered by FLUX.2 Klein, Sana 2.0, Wan2.1, Lumina 2.0 & SD 3.5. Zero cloud, zero API keys, zero fees.

First native Apple Silicon (MLX) port of Diffusion Policy (RSS 2023 Best Paper). 6 policy variants, 472 tests, Metal GPU verified. Train and run visuomotor diffusion policies on M-series — no CUDA required.

Modular LLM inference engine in Rust + Zig SIMD kernels. Runs on desktop (Metal GPU), browser (WASM), and ESP32. INT8/Q4 quantization, speculative decoding, multi-model support.

Distinguished-point Pollard kangaroo solver for Bitcoin puzzle #135, tuned for Apple Silicon CPU

A Ruby C extension providing native macOS window management and Metal GPU acceleration for graphics applications.

Comprehensive VAE performance benchmark comparing PyTorch vs TensorFlow on Apple Silicon (M1/M2/M3). Quantifies training speed, memory efficiency, and Metal GPU utilization across Python versions to guide framework selection for ML prototyping and production deployment.

LeRobot-MLX: HuggingFace LeRobot ported to Apple MLX for native Apple Silicon robotics policy training & inference. 10 policies, 739+ tests, Metal GPU accelerated.