Closed DhanshreeA closed 1 month ago
Hi @DhanshreeA
A lot of work has been done in that front. Please update this issue with all the changes and new features before closing it when it is complete.
Ersilia now has the functionality to both figure out which way a model is packed, or be forced to fetch a model in a way that it is packed with FastAPI, or BentoML.
The resolution happens by way of a TemplateResolver class where Ersilia inspects the file structure of the given model to determine whether to use FastAPI (aka, the new template + ersilia pack format), or to use BentoML (aka, the legacy template + bentoml format) to pack a model while fetching it.
Alternatively, users can also make this decision for ersilia by setting the following flags: --with_fastapi
, or --with_bentoml
.
I will close this as completed but documentation is still pending.
Summary
Ersilia CLI needs to evolve with the new packing strategy and use the correct one when fetching and serving a model.
Objective(s)
Documentation
No response