Fetching the paper…

Predictive Pipelined Decoding: A Compute-Latency Trade-off for Exact LLM Decoding · Around