You shouldn't need changes to the qna.yaml file. Those were part of #3115

You shouldn't need changes to the qna.yaml file. Those were part of #3115

Thanks! I just rebased and it was removed from the PR.

why not just add the --rag flag to ilab model download to have it download the embedding model for the rag use case?

• restricting this behavior to apply only when the feature flag for RAG is turned on

I thought it was covered by the discussion in the related issue about restricting this behavior to apply only when the feature flag for RAG is turned on, but maybe better to ask @jwm4 for confirmation.

I still prefer the approach I specified in the issue:

There was some discussion of restricting this behavior to apply only when the feature flag for RAG is turned on. However, this is problematic if users first run ilab model download with default settings and then turn on the feature flag for RAG and then try to run RAG with default settings. Also, having conditional default values for parameters is just inherently complicated and confusing. Furthermore, the model doesn't take up a huge amount of disk space (under 0.5 GB) so it doesn't seem like a big problem is some users wind up downloading an extra model that they don't wind up using.

However, I understand it is debatable, and I am happy to reconsider if there is a broad consensus that this is not the right approach.