Skip to main content

SGLang PD-Disaggregated Deployment

Deploy prefill/decode disaggregated inference with SGLang.

apiVersion: rbgs.sgl-project.dev/v1
kind: RoleBasedGroup
metadata:
  name: sglang-pd
spec:
  roles:
    - name: prefill
      workloadType: StatefulSet
      replicas: 2
      roleSpec:
        # Prefill configuration
    - name: decode
      workloadType: LeaderWorkerSet
      replicas: 4
      roleSpec:
        # Decode configuration

rbg-pd

View the complete example: SGLang PD-Disagg