dflash-mlx-speculative-decoding

Lossless DFlash speculative decoding for MLX on Apple Silicon — 1.7–4x faster LLM inference using block diffusion drafting with target model verification.

Independent GitHub snapshot · Automated classification · Original documentation governs