antirez Wrote a MiniMax H3 Inference Engine in C for Apple Silicon and Took 898 Stars in Two Days

GitHub·high signalXBlueskyLinkedInCopy link

Redis creator Salvatore Sanfilippo (antirez) published h3.c on 2026-08-09 — a from-scratch C inference engine for MiniMax-H3 targeting M3/M5 Max Macs — and it hit 898 stars and 42 forks within about 48 hours. The README reports a 512×512, 22-frame, 20-step generation at ~16.69s on M5 Max, dropping to ~12.60s (24.5% faster) with token reduction, an 8.02s native 256-square preview, and 19.32s on the int8 path. Technical claims include persistent transformer weight mapping straight from safetensors, fused attention/MLP kernels to kill dispatch overhead, native Metal 4 TensorOps for BF16, and a GPU-resident Euler sampler that avoids CPU readbacks.