The ModelCache hydration Job currently runs python:3.11-slim and pip install huggingface_hub at runtime before hf download. A prebuilt, Modelplane-owned image with the tooling preinstalled would be faster, pinned/reproducible, and remove the network dependency on PyPI at hydration time. It also gives a place to preinstall other source tools (e.g. aws for a future S3 source).
Towards #66.
The ModelCache hydration Job currently runs
python:3.11-slimandpip install huggingface_hubat runtime beforehf download. A prebuilt, Modelplane-owned image with the tooling preinstalled would be faster, pinned/reproducible, and remove the network dependency on PyPI at hydration time. It also gives a place to preinstall other source tools (e.g.awsfor a future S3 source).Towards #66.