magnitudedev / magnitude

Open source inference engine that tunes its kernels on your hardware to find the fastest way to run each model. 2x faster than Ollama. Works on Apple Silicon, NVIDIA, AMD, Intel, or just a CPU.

GKartist75 / Wan2GP-Desktop-Tauri

Wan2GP Desktop Launcher — Windows, Linux and macOS (Tauri) — one-click install, per-GPU kernels, Auto-Tune

poisonxa16 / pxa

The set-and-forget LLM engine for Pascal and Volta: PXQ codec + kernels, auto-tuned per card. Ready-to-run PXQ models in MODELS.md; benchmarks vs llama.cpp in the README.

wq2012 / SpectralCluster

Python re-implementation of the (constrained) spectral clustering algorithms used in Google's speaker diarization papers.