langchain/templates/rag-weaviate/rag_weaviate/chain.py

62 lines
1.9 KiB
Python
Raw Normal View History

2023-10-27 22:19:34 +00:00
import os
from langchain.chat_models import ChatOpenAI
from langchain.document_loaders import WebBaseLoader
from langchain.embeddings import OpenAIEmbeddings
from langchain.prompts import ChatPromptTemplate
from langchain.pydantic_v1 import BaseModel
2023-10-27 22:19:34 +00:00
from langchain.schema.output_parser import StrOutputParser
from langchain.schema.runnable import RunnableParallel, RunnablePassthrough
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.vectorstores import Weaviate
if os.environ.get("WEAVIATE_API_KEY", None) is None:
raise Exception("Missing `WEAVIATE_API_KEY` environment variable.")
if os.environ.get("WEAVIATE_ENVIRONMENT", None) is None:
raise Exception("Missing `WEAVIATE_ENVIRONMENT` environment variable.")
WEAVIATE_INDEX_NAME = os.environ.get("WEAVIATE_INDEX", "langchain-test")
### Ingest code - you may need to run this the first time
# Load
loader = WebBaseLoader("https://lilianweng.github.io/posts/2023-06-23-agent/")
data = loader.load()
# # Split
text_splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=0)
all_splits = text_splitter.split_documents(data)
# # Add to vectorDB
# vectorstore = Weaviate.from_documents(
# documents=all_splits, embedding=OpenAIEmbeddings(), index_name=WEAVIATE_INDEX_NAME
# )
# retriever = vectorstore.as_retriever()
2023-10-27 22:19:34 +00:00
vectorstore = Weaviate.from_existing_index(WEAVIATE_INDEX_NAME, OpenAIEmbeddings())
retriever = vectorstore.as_retriever()
# RAG prompt
template = """Answer the question based only on the following context:
{context}
Question: {question}
"""
prompt = ChatPromptTemplate.from_template(template)
# RAG
model = ChatOpenAI()
chain = (
RunnableParallel({"context": retriever, "question": RunnablePassthrough()})
| prompt
| model
| StrOutputParser()
2023-10-29 22:50:09 +00:00
)
# Add typing for input
class Question(BaseModel):
__root__: str
chain = chain.with_types(input_type=Question)