Documents not being retrieved from persisted database

[jgrabovac]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

from langchain_openai import AzureOpenAIEmbeddings
from langchain_chroma import Chroma

embedding = AzureOpenAIEmbeddings(model=model, azure_deployment="foo")
persist_directory = './docs/chroma/'
vectordb = Chroma(embedding_function=embedding, persist_directory=persist_directory)
print(vectordb.similarity_search("Test query here"))

Description

I already have a vectorstore in my local directory. Before I was using langchain_community to access Chroma but I have switched over to langchain_chroma once I found that the former was deprecated.

I'm trying to call the existing vectorstores that I have and query them, but I'm not getting any output at all.

I've individually tested the embedding model and have also tested commands such as vectordb._collection.count() which both return non-zero values. I am using an embedding and llm model from Azure and the connection there is also working.

On VS Code I do get an error in the terminal of "Command executed x minutes ago, took y seconds and failed (Exit Code 1)". I wasn't getting this before when I was accessing Chroma from langchain_community and I'm not sure what I am doing wrong.

System Info

Current versions:
langchain==0.2.15
langchain-chroma==0.1.3
langchain-community==0.2.7
langchain-core==0.2.37
langchain-openai==0.1.23
langchain-text-splitters==0.2.2

Python version:
3.12.4

Platform:
Windows

[dosubot[bot]]

Hey @jgrabovac! I'm here to help you with any issues you're encountering. I'm a bot designed to assist with bug fixes, answer questions, and guide potential contributors. Let's tackle this together!

I found some similar discussions and issues that might help you troubleshoot the problem:

1. Empty Embeddings property when returning from Indexing a document: This discussion suggests that if embedding generation works correctly outside the index function and the API call returns a 200 OK, the issue might lie in how these embeddings are handled or stored by the index function within the Chroma vector store. It recommends reviewing the integration with Chroma's upsert method and enhancing error handling and logging around the point where embeddings are passed to the upsert method ^[1].

2. Context missing when using Chroma with persist_directory and embedding_function: This discussion suggests ensuring that the documents are correctly loaded and stored in the vector store. It provides an example of how to load documents and store vectors locally, and then load the vector store with persisted vectors ^[2].

Additionally, here are some steps to troubleshoot your issue:

1. Ensure Proper Document Loading and Index Creation:

   • Make sure that the documents are correctly loaded and split before adding them to the vector store.
   • Verify that the index is created with the appropriate parameters.

   Example code for loading documents and creating the index:

   from langchain_core.documents import Document
   from langchain_text_splitters import CharacterTextSplitter
   
   loader = TextLoader(SOURCE_FILE_NAME)
   documents = loader.load()
   text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
   docs = text_splitter.split_documents(documents)
   
   vectorstore = Chroma(
       embedding_function=embedding,
       persist_directory=persist_directory
   )
   vectorstore.add_documents(docs)

2. Check for Proper Initialization of Chroma Collection:

   • Ensure that the Chroma collection is properly initialized and that the documents are correctly added to the collection.

   Example code for adding documents to a Chroma vector store:

   from langchain_chroma import Chroma
   from langchain_openai import OpenAIEmbeddings
   from langchain_core.documents import Document
   
   vector_store = Chroma(
       collection_name="foo",
       embedding_function=OpenAIEmbeddings(),
       persist_directory=persist_directory
   )
   
   document_1 = Document(page_content="foo", metadata={"baz": "bar"})
   document_2 = Document(page_content="thud", metadata={"bar": "baz"})
   documents = [document_1, document_2]
   ids = ["1", "2"]
   vector_store.add_documents(documents=documents, ids=ids)

3. Additional Configuration Steps:

   • When initializing the Chroma class with an existing vectorstore, ensure that you are using a PersistentClient and properly configuring the Chroma class.

   Example code for initializing with a PersistentClient:

   import chromadb
   
   persistent_client = chromadb.PersistentClient()
   collection = persistent_client.get_or_create_collection("collection_name")
   collection.add(ids=["1", "2", "3"], documents=["a", "b", "c"])
   
   vector_store_from_client = Chroma(
       client=persistent_client,
       collection_name="collection_name",
       embedding_function=embedding,
       client_settings=chromadb.config.Settings(is_persistent=True),
       persist_directory=persist_directory
   )

By following these steps and ensuring proper document loading, index creation, and initialization of the Chroma collection, you should be able to resolve the issue with documents not being retrieved from the persisted database ^[3][4][5].

To continue talking to Dosu, mention @dosu.

[jgrabovac]

This is what worked for me:

embedding = AzureOpenAIEmbeddings(model=model, azure_deployment="foo")
persist_directory = './persistdocs/'
persistent_client = chromadb.PersistentClient(path=persist_directory)
vectordb = Chroma(client=persistent_client, 
    collection_name="persist", 
    embedding_function=embedding, 
    client_settings=chromadb.config.Settings(is_persistent=True), 
    persist_directory=persist_directory)

Will close the discussion now.