Skip to content

Latest commit

 

History

History
29 lines (20 loc) · 1.76 KB

model-uri.md

File metadata and controls

29 lines (20 loc) · 1.76 KB

Model URI

Supported URIs

Type Format Note
Google Cloud Storage gs://mybucket/path/to/my/model Please make sure your google cloud storage bucket is public.
PHFS phfs:///path/to/my/model This path is mapped to /phfs/path/to/the/model path in the PrimeHub Notebook.
MLflow Model models:/model-name/version It maps to the MLflow Model Registry. The version could be either a version number or the stage of a model.
NFS nfs://1.2.3.4/to/my/model This path is mapped to /nfs/to/my/model in the Model Deployment containers

How does the Model URI work?

If a PrimeHub Deployment runs with a Model URI, it will start a model downloader before a deployment gets ready to serve.

The deployment and its model downloader will share the same volume /mnt/models. When the downloader finished its job, the deployment could load and create a model to serve.

Test GCS Model URI by Docker

In model deployment, we use the gcr.io/kfserving/storage-initializer image to download the model files into the /mnt/models folder.

You can use the following command to test if the download process is work as expected.

docker run -v ${PWD}/models:/mnt/models --rm  \
  gcr.io/kfserving/storage-initializer \
  gs://primehub-models/pytorch/CIFAR10 \
  /mnt/models