ChromaDB 0.4+ is no longer compatible with client config

[gpapp]

System Info

• Python 3.9.13

• langchain-0.0.235-py3-none-any.whl

• chromadb-0.4.0-py3-none-any.whl

Who can help?

No response

Information

• The official example notebooks/scripts
• My own modified scripts

Related Components

• LLMs/Chat Models
• Embedding Models
• Prompts / Prompt Templates / Prompt Selectors
• Output Parsers
• Document Loaders
• Vector Stores / Retrievers
• Memory
• Agents / Agent Executors
• Tools / Toolkits
• Chains
• Callbacks/Tracing
• Async

Reproduction

Steps to reproduce:

1. Create a Chroma store which is locally persisted

store = Chroma.from_texts(
    texts=docs, embedding=embeddings, metadatas=metadatas, persist_directory=environ["DB_DIR"]
)

2. Get the error You are using a deprecated configuration of Chroma. Please pip install chroma-migrate and run chroma-migrate to upgrade your configuration. See https://docs.trychroma.com/migration for more information or join our discord at https://discord.gg/8g5FESbj for help!
3. Suffer

Expected behavior

1. Create locally persisted Chroma store
2. Use Chroma store

The issue:

Starting chromadb 0.40 the chroma_db_impl is no longer a supported parameter, it uses sqlite instead.

Removing the line
chroma_db_impl="duckdb+parquet",
from langchain.vectorstores/chroma.py solves the issue, but the earlier DB cannot be used or migrated.

[dosubot]

Answer generated by a 🤖

Answer

Thank you for bringing this issue to our attention. It seems that the latest version of ChromaDB is no longer compatible with the client config, causing an error when creating a locally persisted Chroma store.

Based on your description, it appears that you have already identified a potential solution by removing the line chroma_db_impl="duckdb+parquet" from the langchain.vectorstores/chroma.py file. This workaround seems to resolve the issue, but it also means that the earlier DB cannot be used or migrated.

We appreciate your effort in finding a solution to this problem. If you would like to contribute your proposed fix to the LangChain project, we encourage you to create a pull request (PR) with your changes. The maintainers will review your contribution and decide if it should be merged into the project.

Thank you for your support and for helping to improve LangChain!

---

This response is meant to be useful, save you time, and share context. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

[dosubot]

Answer generated by a 🤖

Answer

I understand that you're encountering an issue with ChromaDB 0.4+ compatibility in the LangChain Python framework. The error message you're receiving suggests that the ChromaDB configuration you're using is deprecated and needs to be upgraded. You've found a workaround by removing the line chroma_db_impl="duckdb+parquet" from langchain.vectorstores/chroma.py, but this prevents you from using or migrating the earlier database.

To address this issue, one potential solution could be to use a conditional statement to check the version of ChromaDB being used and then apply the appropriate configuration. Here's how you could implement this:

import chromadb

# Get the version of ChromaDB
chroma_version = chromadb.__version__

# Check if the version is 0.4 or later
if float(chroma_version) >= 0.4:
    # Use the new configuration
    _client_settings = chromadb.config.Settings(
        chroma_db_impl="new_configuration",
        persist_directory=persist_directory,
    )
else:
    # Use the old configuration
    _client_settings = chromadb.config.Settings(
        chroma_db_impl="duckdb+parquet",
        persist_directory=persist_directory,
    )

In this code, "new_configuration" should be replaced with the appropriate configuration for ChromaDB 0.4+.

This solution should allow you to use ChromaDB 0.4+ in a backwards compatible way. However, please note that you may still need to handle the migration process as suggested by the error message. This might involve using the chroma-migrate tool or a similar method to upgrade your existing data to the new configuration.

This is just one potential solution and it might need to be adjusted based on your specific code and requirements. If you find this solution helpful and believe it's not covered by existing documentation, I encourage you to make a pull request to update the documentation for the benefit of other users.

Sources

• langchain/vectorstores/chroma.py

---

This response is meant to be useful, save you time, and share context. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

[lightbot21]

Thanks for helping me solve this problem

[kevinknights29]

Thanks for taking quick action on this issue.

I was trying to figure out how to use the Chroma class with this new update, and found out that this code works:

import chromadb
from langchain.embeddings.sentence_transformer import SentenceTransformerEmbeddings
from langchain.vectorstores import Chroma

client = chromadb.PersistentClient(path="./db")
embeddings = SentenceTransformerEmbeddings(
    model_name="all-mpnet-base-v2",
)
vector_db = Chroma(
    client=client,
    collection_name="your-collection-name",
    embedding_function=embeddings,
)
print(f"Documents Loaded: {vector_db._collection.count()}")

Sharing in case someone else needs it.

[aevedis]

Thanks for taking quick action on this issue.

I was trying to figure out how to use the Chroma class with this new update, and found out that this code works:

import chromadb
from langchain.embeddings.sentence_transformer import SentenceTransformerEmbeddings
from langchain.vectorstores import Chroma

client = chromadb.PersistentClient(path="./db")
embeddings = SentenceTransformerEmbeddings(
    model_name="all-mpnet-base-v2",
)
vector_db = Chroma(
    client=client,
    collection_name="your-collection-name",
    embedding_function=embeddings,
)
print(f"Documents Loaded: {vector_db._collection.count()}")

Sharing in case someone else needs it.

Hi @kevinknights29,

But how can I instead save embeddings and docs into into? Intuitively, I would use the following code:

import chromadb
from langchain.vectorstores import Chroma
from langchain.embeddings.openai import OpenAIEmbeddings
embeddings = OpenAIEmbeddings(openai_api_key = key)

client = chromadb.PersistentClient(path="db_metadata_v5")
vector_db = Chroma(
    client=client,
    embedding_function=embeddings,
)
vector_db = Chroma.from_documents(documents=chunks, embedding=embeddings, persist_directory=output_dir)
vector_db.persist()

But now, according to this, the function is no longer supported. So the final two lines throw an error, and indeed, I have "TypeError: 'NoneType' object is not iterable".
Have you found a solution to this as well?

[maheeppartap]

@aevedis vector_db = Chroma.from_documents(documents=chunks, embedding=embeddings, persist_directory=output_dir) should now be db = vector_db.from_documents(documents=chunks, embedding=embeddings, persist_directory=output_dir) instead, otherwise you are just overwriting the vector_db variable.
also vector_db.persist() is no longer needed and has been removed. It should be implicitly called when you call from_documents.

[aevedis]

db = vector_db.from_documents(documents=chunks, embedding=embeddings, persist_directory=output_dir)

@maheeppartap Eh, but it still does not work. I get TypeError: 'NoneType' object is not iterable

@jeffchuber, maybe we need something on langchain side? Do you have any proposals? :)

[lightbot21]

 I also meet this problem， so I can only installed the older version(chromadb==0.3.21) to solve that.

…

---Original--- From: "Andrey E. ***@***.***> Date: Wed, Jul 19, 2023 23:29 PM To: ***@***.***>; Cc: ***@***.******@***.***>; Subject: Re: [hwchase17/langchain] ChromaDB 0.4+ is no longer compatible withclient config (Issue #7887) db = vector_db.from_documents(documents=chunks, embedding=embeddings, persist_directory=output_dir) Eh, but it still does not work. I get TypeError: 'NoneType' object is not iterable . — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>

[kevinknights29]

Thanks for taking quick action on this issue.
I was trying to figure out how to use the Chroma class with this new update, and found out that this code works:

import chromadb
from langchain.embeddings.sentence_transformer import SentenceTransformerEmbeddings
from langchain.vectorstores import Chroma

client = chromadb.PersistentClient(path="./db")
embeddings = SentenceTransformerEmbeddings(
    model_name="all-mpnet-base-v2",
)
vector_db = Chroma(
    client=client,
    collection_name="your-collection-name",
    embedding_function=embeddings,
)
print(f"Documents Loaded: {vector_db._collection.count()}")

Sharing in case someone else needs it.

Hi @kevinknights29,

But how can I instead save embeddings and docs into into? Intuitively, I would use the following code:

import chromadb
from langchain.vectorstores import Chroma
from langchain.embeddings.openai import OpenAIEmbeddings
embeddings = OpenAIEmbeddings(openai_api_key = key)

client = chromadb.PersistentClient(path="db_metadata_v5")
vector_db = Chroma(
    client=client,
    embedding_function=embeddings,
)
vector_db = Chroma.from_documents(documents=chunks, embedding=embeddings, persist_directory=output_dir)
vector_db.persist()

But now, according to this, the function is no longer supported. So the final two lines throw an error, and indeed, I have "TypeError: 'NoneType' object is not iterable". Have you found a solution to this as well?

---

Based on: https://github.com/hwchase17/langchain/blob/master/langchain/vectorstores/chroma.py

You could:

import chromadb
from langchain.vectorstores import Chroma
from langchain.embeddings.openai import OpenAIEmbeddings
embeddings = OpenAIEmbeddings(openai_api_key = key)

client = chromadb.PersistentClient(path="db_metadata_v5")
vector_db = Chroma.from_documents(
     client=client, 
     documents=chunks, 
     embedding=embeddings, 
     persist_directory=output_dir
)
vector_db.persist()

This is possible because the from_documents method has a client argument.

Could you give it a try? I don't see why it wouldn't work.