is the OpenAI API key cached?

[maspotts]

is the OpenAI API key cached?

[whitead]

Yes, the new version pickles the LLM models which contain the openai key used when constructing the object.

You can call docs.update_llms('gpt-3.5-turbo') (or whatever llm you want) to overwrite the existing llms after loading a paperqa object.

[maspotts]

Aha! Of course! I'm actually using that already, but it didn't occur to me to try it here!

So if I overwrite the existing llm then that will also overwrite the (saved) openai API key?

[maspotts]

Hmmm... what I actually want to do is to update the API key (from os.environ['OPENAI_API_KEY'] or as an explicit string, doesn't matter) but keep the existing llm (use the same model). Is there an easy way to do that? eg. I might have specified 'gpt-4' when I built the index, but the API key has since hit its quota, so now (at query time) I want to load the index from the pickled file and update the API key but continue to use 'gpt-4' (but ideally without having to separately save the string 'gpt-4'). Is that possible?

[whitead]

Yea, just do docs.update_llms('gpt-4'). Make sure you're on version 2 (since version 3 is incompatible with previously pickled docs)

[maspotts]

Ok but what if the index was previously built with a different LLM and I want to continue using the same LLM but just update the API token?MikeOn Jun 20, 2023, at 1:01 PM, Andrew White ***@***.***> wrote: Yea, just do docs.update_llms('gpt-4'). Make sure you're on version 2 (since version 3 is incompatible with previously pickled docs) —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you authored the thread.Message ID: ***@***.***>