Open taneem-ibrahim opened 1 year ago
Should this be a REST or gRPC based API? I would think REST + JSON would be simpler and easier to use and we could probably add the necessary code in the rest-proxy
repository.
@taneem-ibrahim -- Why would we need a swagger UI for each serving runtime instead of relying on modelmesh to route to any replica of the model?
Yes Rest+JSON should be fine. Relying on modelmesh to route is okay too. Thank you!
Is your feature request related to a problem? If so, please describe. Currently the kserve inference swagger UI does not work with model mesh. The swagger UI is very useful to developers because it allows interacting with the APIs in a visual way on the browser.
In case of model mesh we need a way to provide a swagger UI for each serving runtime so that each inference request are routed to right set of replicas for a given model.