community: add milvus hybrid search retriever

[BuxianChen]

Description:
I add milvus hybrid search retriever. The PR messages remains to complete, and I need some advices for the implementation. The code that I already committed can run successfully, but need to be improved, refactor, lint (I have some problem in format, lint, etc).

Here are my plans, and need some helps:

1. May refactor the code, need some advises. Meanwhile, I need do more tests, I believe there are many bugs in my current implementation.
2. Pass the lint, format, mypy checks, add docstrings and type hints.
3. Add more functional: support hybrid search, single vector search and query (i.e. no vector sematic search) by MilvusHybridSearchRetriever.

Dependencies:

Milvus>=2.4.0
pymilvus>=2.4.0

Docs

• an example notebook showing its use. It lives in docs/docs/integrations/retrievers/milvus_hybrid_search.ipynb.

Thank you for contributing to LangChain!

• PR title: "package: description"

  • Where "package" is whichever of langchain, community, core, experimental, etc. is being modified. Use "docs: ..." for purely docs changes, "templates: ..." for template changes, "infra: ..." for CI changes.
  • Example: "community: add foobar LLM"

• PR message: Delete this entire checklist and replace with

  • Description: a description of the change
  • Issue: the issue # it fixes, if applicable
  • Dependencies: any dependencies required for this change
  • Twitter handle: if your PR gets announced, and you'd like a mention, we'll gladly shout you out!

• Add tests and docs: If you're adding a new integration, please include

  1. a test for the integration, preferably unit tests that do not rely on network access,
  2. an example notebook showing its use. It lives in docs/docs/integrations directory.

• Lint and test: Run make format, make lint and make test from the root of the package(s) you've modified. See contribution guidelines for more: https://python.langchain.com/docs/contributing/

Additional guidelines:

• Make sure optional dependencies are imported within a function.
• Please do not add dependencies to pyproject.toml files (even optional ones) unless they are required for unit tests.
• Most PRs should not touch more than one package.
• Changes should be backwards compatible.
• If you are adding something to community, do not re-import it in langchain.

If no one reviews your PR within a few days, please @-mention one of baskaryan, efriis, eyurtsev, hwchase17.

[vercel]

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

Name	Status	Preview	Comments	Updated (UTC)
langchain	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	May 1, 2024 10:24am

[ohadeytan]

Hi, that looks very good @BuxianChen!

As part of a group in IBM Research, we are looking for similar needs: building and using a Milvus vector store with the ability to do hybrid search. We would like to continue to use langchain rather than using pymilvus directly.

@efriis can you share what is the intention with this and similar PRs related to hybrid search?
I saw that under the partners directory you have a HybridSearchRetriever, but it lacks the creation of the collection and similar setting that exists in this PR and in regular MilvusVertorStore.

We are willing to implement and submit PRs with your guidance.

[BuxianChen]

@ohadeytan Hi, I think this PR is out of date, maybe as well as langchain_community/retrievers/milvus.py, maybe all things related to Milvus should be placed to partner directory.

My PR was completed before their partner PR, but theirs has been merged. But as you point out, their implement lack of the creation of the collection and similar setting in regular MilvusVertorStore.

I think you can communicate with langchain's core dev, then integrate the missing part to partner directory. I think this work needs some refactor, as I borrowed a lot code from MilvusVertorStore.

Best wishes!

[BuxianChen]

By the way, I think the awkward things are:

• BaseVectorStore is assumed to with a dense vector embedding.
• BaseRetriever has no abstract interface like add_document.

I'm also confused about how to deal with that.

[ohadeytan]

@BuxianChen, yeah, it seems they are moving to the partners directory, but the question remains, did they support this kind of changes and can provide feedback and guidance.

[zc277584121]

Thank you for you contribution.
Vector store may refer to "dense vector store", and maybe sparse and hybrid functions need to be placed under Retriever. There is now an implementation of MilvusCollectionHybridSearchRetriever.
https://python.langchain.com/v0.2/docs/integrations/retrievers/milvus_hybrid_search/
Here what you see is

collection = Collection(
    ...
)
retriever = MilvusCollectionHybridSearchRetriever(
    collection=collection,
    ...
)

In the near future, MilvusClient SDK will support hybrid.
The best ideal implementation will be like this:

from pymilvus import MilvusClient

client = MilvusClient("milvus_demo.db")

MilvusHybridSearchRetriever(
    client=client,
    ...
)

[efriis]

closing because should be in partner package! seems like it might already be there too