Abstract:Autoregressive decoding is bottlenecked by its sequential nature. Speculative decoding has become a standard way to accelerate inference by using a fast draft model to predict upcoming tokens from a slower target model, and then verifying them in parallel with a single target model forward pass. However, speculative decoding itself relies on a sequential dependence between speculation and verification. We introduce speculative speculative decoding (SSD) to parallelize these operations. While a verification is ongoing, the draft model predicts likely verification outcomes and prepares speculations pre-emptively for them. If the actual verification outcome is then in the predicted set, a speculation can be returned immediately, eliminating drafting overhead entirely. We identify three key challenges presented by speculative speculative decoding, and suggest principled methods to solve each. The result is Saguaro, an optimized SSD algorithm. Our implementation is up to 2x faster than optimized speculative decoding baselines and up to 5x faster than autoregressive decoding with open source inference engines.
供需大变局最近几年，中国钢铁行业经历了一场深刻的需求结构变革。
，更多细节参见谷歌浏览器下载
�@�����ς����A���܂��܂ȃL�����y�[�����X�V���ꂽ�B�����y�V�y�C���y�V�|�C���g�Ŏ��{���Ă����L�����y�[�����܂Ƃ߂��B�Ȃ����ɋL�ڂ̂Ȃ����艿�i�͑S�Đō��݂ŁA�I�������͖����B
Международное агентство по атомной энергии (МАГАТЭ) не выявило в Иране признаков наличия ядерного оружия, однако действующие ограничения на работу инспекторов не дают возможности однозначно подтвердить мирный характер ядерной программы республики. Об этом заявил генеральный директор организации Рафаэль Гросси в соцсети X.