Problem
KServe now has a first-class generative serving CRD, LLMInferenceService, but models-web-app only has first-class support for InferenceService and InferenceGraph.
This leaves LLM users dependent on kubectl for basic visibility into:
- model URI/name
baseRefs composition
- single-node vs prefill/decode vs multi-node topology
- parallelism and scaling
- Gateway/HTTPRoute/Ingress exposure
- scheduler / InferencePool integration
Scope
- Add backend GET routes for:
- list
LLMInferenceService objects in one namespace
- get single
LLMInferenceService
- get related events
- Add frontend types and service methods for
LLMInferenceService
- Add routes/pages for:
- Surface at least:
- name, namespace, ready state, URLs
- model
uri and name
baseRefs
- topology summary:
- single-node
- multi-node (
worker)
- disaggregated (
prefill)
- parallelism
- scaling / replicas
- router mode:
- managed route/gateway
- BYO refs
- scheduler enabled / disabled
Acceptance Criteria
Follow up issue --> #176
Problem
KServe now has a first-class generative serving CRD,
LLMInferenceService, butmodels-web-apponly has first-class support forInferenceServiceandInferenceGraph.This leaves LLM users dependent on
kubectlfor basic visibility into:baseRefscompositionScope
LLMInferenceServiceobjects in one namespaceLLMInferenceServiceLLMInferenceServiceuriandnamebaseRefsworker)prefill)Acceptance Criteria
LLMInferenceServiceresources in the selected namespaceLLMInferenceServicev1alpha1andv1alpha2)Follow up issue --> #176