Device-cloud speculative decoding reduces server computation, but growing a deep draft tree with a small language model (SLM) requires repeated passes on a m…
机构:三星
来源:arXiv 2610.21436 | AI4Papers 论文推荐平台