I have used lanchain, vector database(weaviate/FAISS) and chain (RetrievalQAWithSourcesChain)
from langchain.vectorstores.weaviate import Weaviate
from langchain.llms import OpenAI
from langchain.chains import RetrievalQAWithSourcesChain
import weaviate
from langchain.prompts.prompt import PromptTemplate
from langchain.vectorstores import FAISS
from langchain.embeddings import OpenAIEmbeddings
# API Key needs to be passed in playground
OPEN_API_KEY="sk-xxxxx"
client = weaviate.Client(
url="https://xxxxx.weaviate.network",
additional_headers={
"X-OpenAI-Api-Key": OPEN_API_KEY
}
)
vectorstore = Weaviate(client, "Products", "description")
# vectorstore = FAISS.load_local(
# "./working_fas",
# OpenAIEmbeddings(openai_api_key=OPEN_API_KEY)
# )
llm = OpenAI(model_name="text-davinci-003", temperature=0,
max_tokens=200, openai_api_key=OPEN_API_KEY)
template = """
Return product and price information
--------------------
{summaries}
"""
prompt = PromptTemplate(
input_variables=["summaries"],
template=template,
)
chain = RetrievalQAWithSourcesChain.from_chain_type(llm=llm, retriever=vectorstore.as_retriever(),
return_source_documents=False,
chain_type_kwargs = {"prompt": prompt}
)
result = chain("suggest me an watch", return_only_outputs=True)
print(result)
With FAISS, I am getting some result
With Weaviate, I am getting ValueError
Here is my schema + code on how i insert datas in class
# Define class and property definitions for products
class_def = {
"class": "Products",
"description": "Products",
"properties": [
{
"dataType": ["text"],
"description": "product category",
"name": "category"
},
{
"name": "sku",
"description": "product sku",
"dataType": ["text"]
},
{
"dataType": ["text"],
"name": "product",
"description": "product name"
},
{
"dataType": ["text"],
"name": "description",
"description": "product description"
},
{
"name": "price",
"dataType": ["number"],
"description": "product price"
},
{
"name": "breadcrumb",
"dataType": ["text"],
"description": "product breadcrumb"
},
{
"name": "source",
"dataType": ["text"],
"description": "product url",
},
{
"name": "money_back",
"dataType": ["boolean"],
"description": "money_back / refund available for the product"
},
{
"name": "rating",
"dataType": ["number"],
"description": "product rating"
},
{
"name": "total_reviews",
"dataType": ["int"],
"description": "product total_reviews"
},
{
"name": "tags",
"dataType": ["text"],
"description": "product tags"
},
{
"name": "type",
"dataType": ["text"],
"description": "product type"
}
],
"vectorizer": "text2vec-openai",
}
# Create Class
client.schema.create_class(class_def)
# Insert datas into class
import pandas as pd
import time
df = pd.read_csv("testing.csv")
print(len(df))
for index,row in df.iterrows():
time.sleep(1)
properties = {
"category": row["category"],
"sku": row["sku"],
"product": row["product"],
"description": row["description"],
"price": row["price"],
"breadcrumb": row["breadcrumb"],
"source": row["source"],
"money_back": row["money_back"],
"rating": row["rating"],
"total_reviews": row["total_reviews"],
"tags": row["tags"],
"type": row["type"],
}
print(properties)
client.data_object.create(properties, "Products")
time.sleep(1)
Tried adjusting the prompt!
No luck
I found the answer with some help from the langchain community. To add metadata you must initialise the weaviate vectorstore with attributes=["source"] parameter
vectorstore = Weaviate(client, "Products", "description", attributes=["source"])