EdgeRunner: A Local LLM Inference Engine for Apple Silicon Implemented Purely in Swift

Introduction / Main Floor: EdgeRunner: A Local LLM Inference Engine for Apple Silicon Implemented Purely in Swift

EdgeRunner is a local large language model (LLM) inference engine built entirely with Swift and Metal, optimized specifically for Apple Silicon. It supports direct loading of GGUF format models without conversion or C++ dependencies, achieving a decoding speed of over 230 tokens per second on the M3 Max.