Currently besides the pip_depencies list, BentoML also automatically figures out other local python files that your prediction service need. So if your custom sklearn transformer code is being imported, and even if it’s not published in PyPI, BentoML is still able to package those