Weaviate Hybrid Retriever issue in Langchain for custom vectors

Support
1

Hi,

I am trying to use WeaviateHybridSearchRetriever from Langchain and getting this error when I query:

image
image1692×439 25.4 KB

I am using custom vectors. This is my code

# For using WCS

import weaviate
import json
import os
from langchain.vectorstores import Weaviate
from langchain.llms import OpenAI
from langchain.chains import ChatVectorDBChain
from langchain.embeddings.huggingface import HuggingFaceEmbeddings
from langchain.document_loaders import PyPDFDirectoryLoader
from langchain.embeddings.huggingface import HuggingFaceEmbeddings
from langchain.text_splitter import TokenTextSplitter, CharacterTextSplitter
from langchain.retrievers.weaviate_hybrid_search import WeaviateHybridSearchRetriever
from langchain.schema import Document
import numpy as np
import json

WEAVIATEURL =

client = weaviate.Client(
url = WEAVIATEURL, # Replace with your endpoint
auth_client_secret=weaviate.AuthApiKey(api_key=API_KEY), # Replace w/ your Weaviate instance API key
)

meta_info = client.get_meta()
print(json.dumps(meta_info, indent=2))
client.schema.get()

###############################################################################################

pdf_path = dir_of_pdfs
loader = PyPDFDirectoryLoader(pdf_path)
documents = loader.load()

text_splitter = TokenTextSplitter(chunk_size=500, chunk_overlap=25)
docs = text_splitter.split_documents(documents)
print (f’Now you have {len(docs)} documents’)

embed_model_id = ‘sentence-transformers/all-MiniLM-L6-v2’

embed_model = HuggingFaceEmbeddings(
model_name=embed_model_id,
model_kwargs={‘device’: “cpu”},
encode_kwargs={‘device’: “cpu”, ‘batch_size’: 32}
)

doc_func_source = lambda x: x.metadata[‘source’]
doc_func_content = lambda x: x.page_content
content_docs = list(map(doc_func_content, docs))
source_docs = list(map(doc_func_source, docs))

embeddings = embed_model.embed_documents(content_docs)
print (len(embeddings))
print (len(embeddings[0]))

############################################################################################################

#writing my vectors in JSON

Format the vectors in the desired JSON structure

combined_data = [{“Content”: content, “Source”: source, “Vector”: vector}
for content, source, vector in zip(content_docs, source_docs, embeddings)]

Dump the formatted vectors to a JSON file

with open(‘data.json’, ‘w’) as json_file:
json.dump(combined_data, json_file)

#############################################################################################################

#creating class

if client.schema.exists(“Ragdocs”):
client.schema.delete_class(“Ragdocs”)

schema for class

class_obj = {
‘class’ :‘Ragdocs’,
“vectorizer”: “none” # If set to “none” you must always provide vectors yourself. Could be any other “text2vec-*” also.

}

client.schema.create_class(class_obj) # returns null on success

####################################################################################################################

#importing data

Read and load the JSON data from the file

with open(‘data.json’, ‘r’) as file:
data = json.load(file)

Configure batch process

client.batch.configure(batch_size=1)

Configure a batch process

with client.batch as batch:
# Batch import all Questions
for i, d in enumerate(data):
print(f"importing document no: {i+1}")

    properties = {
        "text": d["Content"],
        "source": d["Source"],            
    }
   
    custom_vector = d["Vector"]  


    client.batch.add_data_object(
        properties,
        "Ragdocs",
        vector=custom_vector
    )

##########################################################################################################################

retriever = WeaviateHybridSearchRetriever(
client=client,
index_name=“Ragdocs”, #index name is class name
text_key=“text”,
attributes=,
create_schema_if_missing=True,
)

code ran without any issue until here

################################################################################################################################
retriever.get_relevant_documents(“what is OpenVINO?”)

From the error it looks like I need to provide vectors for the query separately as I am not using the vectorizer module of weaviate? How do I provide vectors for the query through the retriever?

2

Hi!

Have you seen this python notebook?

I believe that you need to initiate Weaviate client passing the OpenAI API Key so it can vectorize your query.

here:

import os

import weaviate

WEAVIATE_URL = os.getenv("WEAVIATE_URL")
auth_client_secret = (weaviate.AuthApiKey(api_key=os.getenv("WEAVIATE_API_KEY")),)
client = weaviate.Client(
    url=WEAVIATE_URL,
    additional_headers={
        "X-Openai-Api-Key": os.getenv("OPENAI_API_KEY"),
    },
)

# client.schema.delete_all()

Also, you will probabyl find this recipe interesting :wink:

Let me know if that helps :slight_smile:

3

I found this page after having the same issue. I instantiated a WeaviateHybridSearchRetriever retriever but there isn’t a place (as far as I can tell) where I can include the vector representation of the user query. Hence the error. I think this is more on the langchain side than the Weaviate side. BTW, @DudaNogueira your answer to the original issue on this page, created by @Tsan won’t work because you suggested that he pass an OpenAI API key to use the ada embedding model, but all of his vectors on his Weaviate cluster were made using the SentenceTransformers all-MiniLM-L6-v2 model.

4

Hi @americanthinker !!

I believe you may be able to pass the vector using the hybrid_search_kwargs, according to here

something along the lines of (I have not test):

considering this python notebook

retriever.get_relevant_documents("the ethical implications of AI", hybrid_search_kwargs={"vector": [x,y,z]} )

Let me know if that works :slight_smile: