Community Challenge Speeds Up Open-Source Bonsai 2 9B Sixfold

A one-week community optimization challenge sped up the open-source Bonsai 2 9B model by roughly six times, reaching about 959 tokens/s on Apple Silicon Macs and gaining 421% on NVIDIA DGX Spark, aided by tuning of a custom dflash2 runner with speculative decoding.

2026-10-21 ~ 2026-10-23 · 2 related posts