Skip to content

[Feature]: Model Override #221

Description

@jtechapps

Problem Statement

Based of #205

I want to be able to fulfill requests specifying one model with another model instead of having those requests error when sent to the backend/IGW.

Proposed Solution

model_name_override can be specified for a Pool object.

If model_name_override is not empty, every request processed will have the model name replaced in the inference request with the override name. The original name will be stored as internal metadata on the request and be replaced back into the response before placing into the result channel.

Alternatives Considered

Leaving model overriding to inference-gateway backend:
https://gateway-api-inference-extension.sigs.k8s.io/api-types/inferencemodelrewrite/

the problem with this solution is it's incompatible if async-processor is not using IGW. IGW is not a hard dependency for async-processor. Async-processor can be used with model servers directly.

Willingness to Contribute

Yes, I can submit a PR

Additional Context

No response

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions