BentoML

This is not installing tensorflow, your service.py probably include import transformers or diffusers somewhere

if you have tensorflow locally, it will try to import tensorflow

this is a known transformers/diffusers feature

The build step runs the service.py to see which models are referenced by the runners so that it knows which models to package into the bento. a by-product of that is that the imports are run. Which is ok in most cases because your local environment typically has those dependencies because you were developing the bento in that environment. There are also work arounds if you need too