dLLM Notes

Paper notes · ICML 2026

DFlash: Block Diffusion for Flash Speculative Decoding

Jian Chen, Yesheng Liang, Zhijian Liu

Speculative decoding speeds up an AR model: a cheap drafter guesses several tokens, the target model checks them all in one pass, and the output is exactly what the target would have written alone. DFlash keeps the target AR and changes the drafter: instead of a small AR model that guesses one token per pass, a small block diffusion model guesses a whole block of 16 in one pass.

A dLLM is a weak writer on its own (M06, M12). As a drafter it does not need to be good, only often right, because every guess is checked. That is the paper’s reframing.

Three ingredients:

One answer, three ways, in toy cost units (one step of the big model = 1). Classic speculative decoding: a small AR drafter guesses 4 tokens, one pass each. DFlash: a small diffusion drafter guesses a block of 8 in one pass, and the big model checks it. All three output the big model's greedy text.

Loading PDF… open it directly

PDF reproduced unmodified under CC BY 4.0. Highlights and notes are mine.