English

D\'ej\`aVu: KV-cache Streaming for Fast, Fault-tolerant Generative LLM Serving

Distributed, Parallel, and Cluster Computing 2024-03-05 v1

Abstract

Distributed LLM serving is costly and often underutilizes hardware accelerators due to three key challenges: bubbles in pipeline-parallel deployments caused by the bimodal latency of prompt and token processing, GPU memory overprovisioning, and long recovery times in case of failures. In this paper, we propose D\'ej\`aVu, a system to address all these challenges using a versatile and efficient KV cache streaming library (D\'ej\`aVuLib). Using D\'ej\`aVuLib, we propose and implement efficient prompt-token disaggregation to reduce pipeline bubbles, microbatch swapping for efficient GPU memory management, and state replication for fault-tolerance. We highlight the efficacy of these solutions on a range of large models across cloud deployments.

Keywords

Cite

@article{arxiv.2403.01876,
  title  = {D\'ej\`aVu: KV-cache Streaming for Fast, Fault-tolerant Generative LLM Serving},
  author = {Foteini Strati and Sara Mcallister and Amar Phanishayee and Jakub Tarnawski and Ana Klimovic},
  journal= {arXiv preprint arXiv:2403.01876},
  year   = {2024}
}
R2 v1 2026-06-28T15:08:08.766Z