vllm/docs/deployment/integrations/llm-d.md

llm-d

llm-d is a Kubernetes-native distributed inference framework for serving large language models at scale, with vLLM as its primary inference engine. llm-d coordinates a fleet of vLLM instances across a cluster so that performance holds up under real production traffic, achieving the fastest "time to state-of-the-art (SOTA) performance" for key OSS models across most hardware accelerators.

It is a CNCF Sandbox project founded by Red Hat, Google Cloud, IBM Research, CoreWeave, and NVIDIA.

What llm-d adds to vLLM

A single vLLM server is fast, but at scale the picture changes: across many replicas, cache locality breaks under round-robin load balancing, long prompts inflate time-to-first-token, and accelerators sit underused. llm-d adds the cluster-level layer that vLLM does not aim to provide on its own:

• Prefix-aware routing. Instead of round-robin, llm-d reads vLLM's KV-cache events and routes each request to the replica that already holds its prefix, reusing cache instead of recomputing it.
• Distributed KV-cache management. A global index tracks which token blocks live on which replica, and tiered offloading spills cache to CPU memory or local SSD, extending the working set beyond accelerator HBM.
• Prefill/decode disaggregation. Prompt processing and token generation run on separate vLLM workers, with KV-cache moved over the vLLM NIXL connector, lowering TTFT and steadying per-token latency on long prompts.
• Wide expert-parallelism. Serve large Mixture-of-Experts models such as DeepSeek-R1 and GPT-OSS across nodes with combined data and expert parallelism, for more KV-cache capacity and throughput.
• SLO-aware autoscaling and flow control. Scale vLLM pools on real inference signals (queue depth, true demand) rather than raw GPU utilization, with multi-tenant fairness and priority dispatch.

These are composable. Most teams start by adding prefix-aware routing over an existing vLLM pool, then layer in the rest as specific bottlenecks appear.

Performance

Representative benchmarked results across accelerators:

• 3x higher output throughput and 2x faster TTFT from prefix-aware routing vs round-robin (Llama 3.1 70B, AMD MI300X)
• Up to 70% higher tokens/sec from prefill/decode disaggregation (GPT-OSS, NVIDIA B200)
• 13.9x throughput from hierarchical KV offloading at high concurrency vs GPU-only (NVIDIA H100)

See the full list and reproducible benchmarks on Prism.

Get started

1. Deploy the Optimized Baseline with the Quickstart. It stands up an intelligent router over a vLLM pool on Kubernetes in a tested configuration.
2. Browse the well-lit path guides, each a tested recipe for one of the capabilities above, and add the optimization that fits your workload.
3. Read the Introduction and Architecture overview to see how the pieces wrap your vLLM deployment.

You can also deploy vLLM with llm-d via KServe's LLMInferenceService.

Questions and contributions are welcome on GitHub and Slack.