标题： 月饼：面向大语言模型服务的以KV缓存为中心的解耦架构 显示英文标题

标题： Mooncake: A KVCache-centric Disaggregated Architecture for LLM Serving

摘要： 月饼是Kimi的部署平台，Kimi是由Moonshot AI提供的领先大语言模型服务。 它采用以KV缓存为中心的解耦架构，将预填充和解码集群分离。 它还利用GPU集群中未充分利用的CPU、DRAM和SSD资源，实现KV缓存的解耦缓存。 月饼的核心是其以KV缓存为中心的调度器，在最大化整体有效吞吐量的同时满足与延迟相关的服务等级目标（SLO）。 与传统研究假设所有请求都将被处理不同，月饼由于高度过载的场景而面临挑战。 为缓解这些问题，我们开发了一种基于预测的早期拒绝策略。 实验表明，月饼在长上下文场景中表现出色。 与基线方法相比，月饼在某些模拟场景中可以在遵守SLO的前提下实现吞吐量最高提升525%。 在实际工作负载下，月饼的创新架构使Kimi能够处理75%更多的请求。 显示英文摘要

摘要： Mooncake is the serving platform for Kimi, a leading LLM service provided by Moonshot AI. It features a KVCache-centric disaggregated architecture that separates the prefill and decoding clusters. It also leverages the underutilized CPU, DRAM, and SSD resources of the GPU cluster to implement a disaggregated cache of KVCache. The core of Mooncake is its KVCache-centric scheduler, which balances maximizing overall effective throughput while meeting latency-related Service Level Objectives (SLOs). Unlike traditional studies that assume all requests will be processed, Mooncake faces challenges due to highly overloaded scenarios. To mitigate these, we developed a prediction-based early rejection policy. Experiments show that Mooncake excels in long-context scenarios. Compared to the baseline method, Mooncake can achieve up to a 525% increase in throughput in certain simulated scenarios while adhering to SLOs. Under real workloads, Mooncake's innovative architecture enables Kimi to handle 75% more requests.