#欢迎关注爱范儿官方微信公众号：爱范儿（微信号：ifanr），更多精彩内容第一时间为您奉上。
Двое мужчин подошли к пилоту, который только что выпрыгнул с парашютом из подбитого самолета. Один из них размахивал над головой чем-то вроде металлической трубы, приняв, по-видимому, летчика за иранца.，更多细节参见heLLoword翻译官方下载
This week, with global tensions spiking, the Cold Open returned to familiar territory: James Austin Johnson behind a podium, doing his uncanny Trump, explaining why the United States launched attacks on Iran.，推荐阅读PDF资料获取更多信息
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.