I have been looking to use lipsync for a project at work, and I have been horrified to see that the most used deepfake library [0] with over 30k stars has the word "test" mentioned in it a whopping 10 times. This naturally made me very hesitant to use this library as who (I guess no one, really) knows how unstable it may be.<p>Are there any examples of projects in ML (other than libs like pytorch) that do have test suites, and perhaps even enforce TDD? Would be very curious to know.<p>[0] DeepFaceLab https://github.com/iperov/DeepFaceLab
I try to have test coverage for the (small) independent libraries I do.<p>It's a bit tricky to go really deep with tests when you don't have a reference, but I like to see that at least basic logic errors and typos in attribute names are caught.<p>Many of the larger libraries in the PyTorch ecosystem have some sort of testing. The things that are mostly "here is a model" often do not.