Language
HomeLanguages
Sections
MTPLX leverages built-in MTP heads in modern models like Qwen 3.5-3.8 to achieve up to 2.24x speedup on Apple Silicon without requiring a separate draft model.
A practical guide to fine-tuning 8B models on consumer hardware using Soup—a CLI tool that brings layer streaming and automated pipelines to small GPUs.
Skales is a local AI agent for desktop and mobile that runs autonomously on your system, reads files, manages your browser, and executes task chains without constant oversight.
Build a private surveillance camera on Raspberry Pi with end-to-end encryption and zero-trust architecture — no cloud subscriptions required.
Engineer Andrey Mikhailov built TurboFieldfare, a custom Swift/Metal runtime that runs Google's Gemma 4 26B-A4B model in just 2 GB of RAM on a basic MacBook Air M2.