A gitlab repository for research scientist to develop and deploy http services:

• a template project that serves as a starting point. It includes a web server, unit tests, C/I integration, example config etc.
• an example project, documented as a how-to for using the template
• integrated with the wikimedia deployment pipeline to create docker images that are production ready
• tooling for deploying the service(s) to toolforge, cloudVPS and yarn

The purpose of this repository is to eventually replace the usages of the research-api-endpoint-template github project with an alternative on WMF gitlab, and supported by tooling like C/I and the WMF deployment pipeline.

The api endpoints that make use of this template should not be centered on ML inference, but rather on functionality that can't be deployed on liftwing (e.g. proof of concept webpages/ui, rest endpoints doing additional logic besides inference etc.). The boundary between what can/should be deployed on liftwing is not always obvious, this template project will include both 1. documentation/guidlines on how to decide that boundary 2. helpful abstractions to facilitate deploying a part of a proof of concept API/UI as a liftwing service.

Impact:
In the effort of unifying tooling for research scientists, we would like to retired the github project, and create a new gitlab project that serves a larger audience in the research group.

Upon completion of this work, the need for research scientists to code up bespoke wrappers for deploying their project outside of the liftwing ecosystem will be significantly reduce/removed, while the new repo will help upholding engineering standards at the same time.

Having this gitlab repo will also reduce the overhead from research engineering front should we decide to move some functionalities/models to production environment and provide seamless integration point between research and ML. In the meantime, it minimizes the efforts from research engineers to maintain projects across the team.