H3-metal

Overview

A native inference engine for MiniMax-H3 hybrid SSM/attention models, written by Salvatore Sanfilippo and aimed at Apple Silicon. Devs should care because it reflects the push toward fast, local, GPU-light inference for modern models.