Do you have a central codebase with the ML models you use, or do you have multiple repos each with a different model? How do you use Huggingface or author's code bases?
At my company, we have a central codebase with different models. So each time we want to use a new model we refactor it from the open source repo and integrate it into our codebase: this includes changing the layers (e.g. Attention/MLP) to use our implementation rather than the custom author's implementation.
I am curious to see what others' workflow looks like, to see how I can make myself more productive.