core: added FakeVectorStore

[lkuligin]

#9292

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

1 Skipped Deployment

Name	Status	Preview	Comments	Updated (UTC)
langchain	⬜️ Ignored (Inspect)	Visit Preview		Jul 8, 2024 7:41pm

[lkuligin]

@eyurtsev @hwchase17 could you take a look, please?

[leo-gan]

@baskaryan Could you, please, review it? It looks like very helpful addition.

[baskaryan]

out of curiosity how are you imagining using this?

[lkuligin]

@baskaryan I find it useful to use together with a FakeLLM (and a callback). It makes it possible to write unit tests to test custom chains.
So essentially you create a few Documents (with some content and metadata) for testing purposes, add them to a fake vectorstore, and then you can unit test the whole chain's logic (that all prompts are formatted correctly, and your final output / evidence also look nice).

Does it make sense?

[efriis]

@lkuligin if you want to write some unit tests for some of the in-library chains and runnables, it might demonstrate the use case a bit better.

Otherwise, I'll probably close this, and you can continue using it in your codebase for unit testing custom chains!

[leo-gan]

We also have plenty Fake classes: LLM, Embeddings. It makes sense to make Fake classes for all modules. So, I like this Fake vector store.

[nejch]

Thanks for this @lkuligin, I was able to use this in combination with a FakeStreamingListLLM to test incorporating chain building into a django app with a chatbot and custom contexts. So just adding another voice/use case here it might make sense since there's similar fake classes already provided with the package 🙇

[eyurtsev]

Let's move the in memory vectorstore into core! It relies on numpy, but we can make it into an optional dependency?

[efriis]

moved to core!

[eyurtsev]

Here's a PR that should resolve this issue: #23986

[eyurtsev]

Closing this as we merged the in memory vectorstore