Download model artifact with HTTP

[leolmz]

Hi, I want to pass a link for Kserve to download a model registered in Mlflow, which uses an HTTP request method to do that (it can be downloaded directly from GitHub or HuggingFace). Will setting up an artifact store in S3 or other public storage service help, i.e using the runId to get the absolute path in these storage services?

I am assuming for example if we can do some character string operation with the artifact_url we get "mlflow-artifacts:/0/ebbbb937c23449d695f8146c4a8241ff/artifacts/onxx-model" to get the exact path of the onxx file on S3 to skip the client.download_artifacts method?

[leolmz]

Hi, @Retired_mod Thank you so much for the reply.

May I ask about the reference or doc for 'KServe can download the model locally from the MLflow artefact store'? The KServe doc I checked about integration with MLflow is this one: https://kserve.github.io/website/0.10/modelserving/v1beta1/mlflow/v2/#training and it seems to put the model artifact into the third model repository and that is not what I am seeking for. 

[Kumaran]

Hi @leolmz,

You can refer to the doc for downloading the model artifacts