Backend Installation Slow Due to Heavy Dependencies Like PyTorch When Using Only OpenAI Endpoints

[Sinnaeve]

Hi,

On DEV branch, deploying the app locally take a lot of time in particular the backend.
Requirement.txt file from the backend contains packages (sentence-transformers, effdet) that have heavy dependencies such as pytorch for example, but if I use openai LLM and embedding model I'm not sure for what we purpose use those packages.
So my question si two fold:

• why do we need packages such as pytorch in the backend ?
• is there a way to install the backend faster, like selecting a version that doesn't require heavy packages if we only want to use endpoints such as openai for LLM and embedding ?

[kartikpersistent]

#620