Memory (ChromaDB) embedding_function support

[jcroucherMSFT]

Does ChromaDBVectorMemory support embedding_functions such as OpenAIEmbeddingFunction?
I can't see an embedding_function input argument in PersistentChromaDBVectorMemoryConfig, and I'm not sure how else to specify an embedding model such as "text-embedding-3-large"

Thanks!

[junmiron]

I think, I got it working by specifying the embedding_function in the chormadb collection before setting the ChromaDBVectorMemoryConfig, but I have to do more testing. It will be great if would support the embedding_function directly in PersistentChromaDBVectorMemoryConfig

--- EMBEDDING FUNCTION (Local SentenceTransformer) ---

emb_fn = embedding_functions.SentenceTransformerEmbeddingFunction(
model_name="all-distilroberta-v1 "
)

--- CHROMADB SET UP ---

Step 2: Set up ChromaDB client with embedding function

chroma_client = chromadb.Client(Settings(
persist_directory=CHROMA_DIR,
anonymized_telemetry=False
))

Step 3: Create or load the collection WITH embedding

collections = chroma_client.get_or_create_collection(
name=COLLECTION_NAME,
embedding_function=emb_fn
)

--- RAG-ENABLED AUTO-GEN SETUP WITH CHROMADB ---

vector_memory = ChromaDBVectorMemory(
config=PersistentChromaDBVectorMemoryConfig(
collection_name=COLLECTION_NAME,
persistence_path=CHROMA_DIR,
k=3,
score_threshold=0.4,
)
)

[jcroucherMSFT]

I wasn't aware a suggested solution has already been made in #6267

[junmiron]

Wow that's nice, hope it gets included soon!