Skip to content

Inference Graph


ML inference systems are getting bigger and more complex, they often consist of many models to make a single prediction. Some common use cases are image classification and natural language processing pipelines. For example, a face recognition pipeline may need to first locate faces in a image, then compute the features of the faces to match records in a database. An NLP pipeline needs to first run document classification, then perform named entity detection downstream based on the previous classification results.

KServe has unique strengths for building a distributed inference graph: an autoscaling graph router, native integration with individual InferenceServices, and a standard inference protocol for chaining models. KServe leverages these strengths to build an InferenceGraph and enable users to deploy complex ML inference pipelines to production in a declarative and scalable way.



  • InferenceGraph: Made up of a list of routing Nodes, where each Node consists of a set of routing Steps. Each Step can either route to an InferenceService or another Node defined on the graph which makes the InferenceGraph highly composable. The graph router is deployed behind an HTTP endpoint and can be scaled dynamically based on request volume. The InferenceGraph supports four different types of Routing Nodes: Sequence, Switch, Ensemble, Splitter.

  • Sequence Node: Allows users to define multiple Steps with InferenceServices or Nodes as routing targets in a sequence. The Steps are executed in sequence and the request/response from the previous step can be passed to the next step as input based on configuration.

  • Switch Node: Enables users to define routing conditions and select a step to execute if it matches a condition. The response is returned as soon it finds the first step that matches the condition. If no condition is matched, the graph returns the original request.

  • Ensemble Node: A model ensemble requires scoring each model separately and then combining the results into a single prediction response. You can then use different combination methods to produce the final result. Multiple classification trees, for example, are commonly combined using a "majority vote" method. Multiple regression trees are often combined using various averaging techniques.

  • Splitter Node: Allows users to split the traffic to multiple targets using a weighted distribution.


Headers Propagation

If you want Inference Graph's router to propagate the headers, you passed in the request to Inference Graph, to all the steps in your graph then you can do so using inferenceservice-config config-map in kserve namespace. For example: If you want to propagate a certain header, say "Custom-Header", then you can edit the router section of inferenceservice-config config-map like this :

    "image" : "kserve/router:v0.11.0",
    "memoryRequest": "100Mi",
    "memoryLimit": "1Gi",
    "cpuRequest": "100m",
    "cpuLimit": "1",
    "headers": {
Once you update this config-map, kserve controller will automatically reconcile Inference Graph to start propagating headers.

Back to top