DFlash

DFlash brings speculative decoding to Apple Silicon using stock MLX without a fork. It is positioned as a performance-focused inference enhancement for local model execution on Apple hardware.