Is it possible to save Recursive_index in Faiss Vector store locally?

[mraguth]

I am using llama_parse and creating recursive_index on fly. Is it possible to store vectors after parse?

Code Example taken from : https://github.com/run-llama/llama_parse/blob/main/examples/demo_insurance.ipynb

Here my objective is store the vector index locally either in Faiss or ChromaDB and call the stored index and run the query_1 without parsing again to save cost.

Appreciate some code samples to store vectors and retrieve vectors for query.

from llama_index.core.node_parser import MarkdownElementNodeParser

node_parser = MarkdownElementNodeParser(
    llm=OpenAI(model="gpt-3.5-turbo-0125"), num_workers=8
)
nodes = node_parser.get_nodes_from_documents(documents)
base_nodes, objects = node_parser.get_nodes_and_objects(nodes)

recursive_index = VectorStoreIndex(nodes=base_nodes + objects)
query_engine = recursive_index.as_query_engine(similarity_top_k=25)
query_1 = "My trip was delay and I paid 45, how much am I cover for?"

response_1 = query_engine.query(query_1)
print(str(response_1))

[dosubot[bot]]

Yes, it is possible to save a Recursive_index in a Faiss Vector store locally. Below are detailed code samples for both Faiss and ChromaDB to store and retrieve vectors for querying.

Faiss

Storing Vectors

from llama_index.legacy.vector_stores.faiss import FaissVectorStore
import faiss
import numpy as np
from llama_index.legacy.schema import BaseNode

# Create a faiss index
d = 1536  # dimension
faiss_index = faiss.IndexFlatL2(d)

# Initialize the vector store
vector_store = FaissVectorStore(faiss_index=faiss_index)

# Add vectors to the store
vectors = [np.random.rand(d).astype('float32') for _ in range(10)]
nodes = [BaseNode(embedding=vector) for vector in vectors]
vector_store.add(nodes)

# Persist the vector store to disk
persist_path = 'faiss_index.bin'
vector_store.persist(persist_path)

Retrieving Vectors

from llama_index.legacy.vector_stores.faiss import FaissVectorStore
import numpy as np
from llama_index.legacy.vector_stores.types import VectorStoreQuery

# Load the vector store from disk
persist_path = 'faiss_index.bin'
vector_store = FaissVectorStore.from_persist_path(persist_path)

# Query the vector store
query_vector = np.random.rand(1536).astype('float32')
query = VectorStoreQuery(query_embedding=query_vector.tolist(), similarity_top_k=5)
result = vector_store.query(query)

print(result.ids, result.similarities)

ChromaDB

Storing Vectors

import chromadb
from chromadb.config import Settings
import numpy as np

# Initialize ChromaDB client
client = chromadb.Client(Settings(chroma_db_impl="duckdb+parquet", persist_directory="./chromadb"))

# Create a collection
collection = client.create_collection("example_collection")

# Add vectors to the collection
vectors = [list(np.random.rand(1536).astype('float32')) for _ in range(10)]
ids = [str(i) for i in range(10)]
collection.add(embeddings=vectors, ids=ids)

Retrieving Vectors

import chromadb
from chromadb.config import Settings
import numpy as np

# Initialize ChromaDB client
client = chromadb.Client(Settings(chroma_db_impl="duckdb+parquet", persist_directory="./chromadb"))

# Get the collection
collection = client.get_collection("example_collection")

# Query the collection
query_vector = list(np.random.rand(1536).astype('float32'))
results = collection.query(query_embeddings=[query_vector], n_results=5)

print(results)

These code samples demonstrate how to store and retrieve vectors using both Faiss and ChromaDB ^[1][2].

To continue talking to Dosu, mention @dosu.

[mraguth]

@dosu While I am creating Faiss vector in a different directory, I am getting error 'default__vector_store.json' file not found. Please help.

# Persist the vector store to disk
persist_path = 'faiss_index.bin'
vector_store.persist(persist_path)

[dosubot[bot]]

To resolve the error 'default__vector_store.json' file not found when creating a Faiss vector store in a different directory, ensure that the persist_path you provide exists and is correct. The error occurs because the file at the specified path does not exist.

Here's how you can modify the from_persist_dir method to specify a different directory:

@classmethod
def from_persist_dir(
    cls,
    persist_dir: str = DEFAULT_PERSIST_DIR,
    fs: Optional[fsspec.AbstractFileSystem] = None,
) -> "FaissVectorStore":
    persist_path = os.path.join(
        persist_dir,
        f"{DEFAULT_VECTOR_STORE}{NAMESPACE_SEP}{DEFAULT_PERSIST_FNAME}",
    )
    # only support local storage for now
    if fs and not isinstance(fs, LocalFileSystem):
        raise NotImplementedError("FAISS only supports local storage for now.")
    return cls.from_persist_path(persist_path=persist_path, fs=None)

Ensure that the persist_dir you provide contains the default__vector_store.json file. If the file does not exist, you will need to create it or specify the correct path where the file is located.

Example usage:

from llama_index.legacy.vector_stores.faiss import FaissVectorStore

# Specify the directory where 'default__vector_store.json' is located
persist_dir = "/path/to/your/directory"

# Create the FaissVectorStore from the specified directory
vector_store = FaissVectorStore.from_persist_dir(persist_dir=persist_dir)

Make sure that the directory /path/to/your/directory contains the default__vector_store.json file. If the file is not there, you will need to create it or move it to the specified directory ^[1][2].

To continue talking to Dosu, mention @dosu.