Open-source engine kapitsa boosts Mac local LLM inference with tree-based speculative decoding

A developer released kapitsa, a hand-written C++20 + Metal inference engine for Apple Silicon featuring tree-based speculative decoding (kflash). It runs Qwen3.8-27B at 140 tok/s on Mac, twice as fast as Ollama, with 23% higher multi-agent throughput.

2026-10-24 ~ 2026-10-24 · 2 related posts

• kapitsa: hand-written Metal inference engine hits 140 tok/s on Qwen3.8-27B for local OpenAI/Anthropic APIs — DotaMate · 2026-10-24
• Open-source kapitsa engine boosts Mac local agent throughput 23% with tree speculative decoding — DotaMate · 2026-10-24