Fetching the paper…

Distributed Speculative Inference (DSI): Speculation Parallelism for Provably Faster Lossless Language Model Inference · Around