DCOCO introduced as a speculative decoding method

In an Oct. 25 post, @roy_eisenstadt introduced DCOCO, short for Contextual Correction of Parallel Drafts, as a speculative decoding method. The author claims it restores context between predicted tokens without giving up parallelism and is up to 6.6× faster than vanilla decoding. The opening post text does not specify benchmark conditions, tested models or measurements for that speed comparison.