Add Chroma to Vector Database examples

[atroyn]

This PR adds Chroma to the vector database examples in the OpenAI cookbook.

Chroma is a lightweight, developer-friendly embeddings store which aims to get A.I application developers up and running with embeddings as fast as possible.

This PR:

• Adds Chroma as another example to the generic Using_vector_databases_for_embeddings_search.ipynb
• Adds an example which shows some of the different tradeoffs in using document retrieval with an embeddings store for LLM-based document Q/A. This uses the gpt3.5-turbo api as well as the ada-002 embeddings API.
• Adds a scientific question/answering dataset for use with the above example, in jsonl format.

[ted-at-openai]

Thanks for the PR; I'll take a look. Always happy to include more solutions for end users.

[csvoss]

This was a fun read! I feel like I actually understand how the hallucinated abstracts improve the overall performance, thanks to the notes and commentary.

This code is outside my regular scope but I happened to be here looking this notebook over anyways, so I left notes on a few minor spelling nits that I noticed while reading.

[atroyn]

Thanks @csvoss for catching those spelling nits!

[atroyn]

@ted-at-openai - let me know if there are any changes you'd like me to make here!

[ted-at-openai]

I'll take a look - thanks for the ping.

[atroyn]

@ted-at-openai following up here again - i'm concerned that this will grow stale soon, we'd like to land this ahead of the retrieval plugin integration.

[ted-at-openai]

Thanks for following up. Here is my current thinking:

Right now the PR can't be merged because due to merge conflicts stemming from another update made to the notebook.

Ideally I'd like to have one notebook per provider, so we don't have a huge blob of concatenated examples, which is (a) hard to read and (b) hard to have multiple people merge to.

Option 1:

• You can fix the merge conflicts by applying your changes to the updated notebook (sorry for the extra work)

Option 2:

• You can wait for us to disentangle the notebook and try to apply your commits to that new framework (this will take longer).

Option 1 will go fastest, but take more work from your side.

I'm sorry I haven't had the time to do this; unfortunately it's not my at the top of my priority list right now. I still really appreciate your contributions, am happy to merge them as soon as ready, and am happy to continue discussing.

[atroyn]

Happy to resolve the conflict on our end

[atroyn]

All set.

[ted-at-openai]

Merged! Thanks for your patience during the long delay. Appreciate your quick changes. Glad to you have your example in the repo.

[atroyn]

Thanks for reviewing! I know it's a busy time, much appreciated.