Checklist

• 1. If the issue you raised is not a feature but a question, please raise a discussion at https://github.com/sgl-project/sglang/discussions/new/choose Otherwise, it will be closed.
• 2. Please use English, otherwise it will be closed.

Motivation

Mooncake is a distributed KVCache storage engine specifically designed for inference with large language models (LLM) based on Transfer Engine. It is a central component in the KVCache-centric distributed architecture. The goal of Mooncake is to store reusable KV caches at various locations within the inference cluster.

Integrate Mooncake as a role in the RBG-deployed SGLang inference service, providing KVCache offload capabilities for the inference service.

Related resources

sglang-integration