8000 Support multiple StorageUri in Inference Service · Issue #3413 · kserve/kserve · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content
Support multiple StorageUri in Inference Service #3413
Open
@lizzzcai

Description

@lizzzcai

/kind feature

Describe the solution you'd like

Crrently KServe only supports one StorageUri, which fit most of the cases. However, in some scenarios like serving fine-tuned models like LoRA, base model and the LoRA adapters might come from different buckets (it is possible to put it into one but in some cases, it will not work if they belong to different owners). So allowing users to provide multiple storageUri will help in this case, similar to volumeMounts in K8s.

Anything else you would like to add:
[Miscellaneous information that will assist in solving the issue.]
As multiple storageUri are supported, user should be able to specify the destination of the mounted storage. Below is one of the example:

apiVersion: serving.kserve.io/v1beta1
kind: InferenceService
metadata:
  name: "huggingface"
spec:
  predictor:
    model:
      modelFormat:
        name: huggingface
      runtime: kserve-huggingface
      storageUri:
      - src: gs://kfserving-examples/models/huggingface
        dest: /mnt/models
      - src: gs://kfserving-examples/adapters/huggingface
        dest: /mnt/adapters

Links to the design documents:
[Optional, start with the short-form RFC template to outline your ideas and get early feedback.]
[Required, use the longer-form design doc template to specify and discuss your design in more detail]

Metadata

Metadata

Assignees

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions

    0