langchainembeddingdtypechromadbvector-database

Precision used in ChromaDB Index

I am using BAAI/bge-large-en-v1.5 model to embed and then store these embeddings in ChromaDB vector-store. These embeddings are in the memory and using HNSW indexing. Is there a way I can find out the dtype or precision of these embeddings if they are float32, float64 or something else?

Thanks

Solution

  • like this:

    import chromadb

# Initialize the DB
client = chromadb.PersistentClient(path="./chroma_db")  # Adjust the path as needed
collection = client.get_collection("my_collection")

# Get a vector by id
vector_data = collection.get(ids=["your_vector_id"], include=["embeddings"])

# Check dtype
if "embeddings" in vector_data and vector_data["embeddings"]:
    vector_array = np.array(vector_data["embeddings"])  # Convert to NumPy array
    print("Vector dtype:", vector_array.dtype)

    By default, ChromaDB stores vectors as float32