Skip to content

Provided serve file in container should allow override of model repository #8668

@Pwalne

Description

@Pwalne

Is your feature request related to a problem? Please describe.
When running triton in sagemaker, it uses the /usr/bin/serve script as a entrypoint. It seems to be hardcoded to use /opt/ml/model as the model repository. This is fine IF you are using sagemaker to download the triton model. If you try to use the sagemaker model to download the actual model, it will fail as its attempting to download and load a triton ensemble/model instead.

Describe the solution you'd like
Change SAGEMAKER_SINGLE_MODEL_REPO=/opt/ml/model/ to SAGEMAKER_SINGLE_MODEL_REPO=${SAGEMAKER_TRITON_MODEL_REPOSITORY:-/opt/ml/model}

Describe alternatives you've considered
There isn't really any other straight forward alternatives. I do not wish to package the model and triton model inside and have sagemaker download it.

Additional context
There isn't really any reason to not do this, it provides more control with no downside.

Metadata

Metadata

Assignees

No one assigned

    Labels

    EnhancementNew feature or request

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions