langchain/templates/rag-gpt-crawler/rag_gpt_crawler/chain.py

import json
from pathlib import Path

from langchain.schema import Document
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.chat_models import ChatOpenAI
from langchain_community.embeddings import OpenAIEmbeddings
from langchain_community.vectorstores import Chroma
from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.pydantic_v1 import BaseModel
from langchain_core.runnables import RunnableParallel, RunnablePassthrough

# Load output from gpt crawler
path_to_gptcrawler = Path(__file__).parent.parent / "output.json"
data = json.loads(Path(path_to_gptcrawler).read_text())
docs = [
    Document(
        page_content=dict_["html"],
        metadata={"title": dict_["title"], "url": dict_["url"]},
    )
    for dict_ in data
]

# Split
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100)
all_splits = text_splitter.split_documents(docs)

# Add to vectorDB
vectorstore = Chroma.from_documents(
    documents=all_splits,
    collection_name="rag-gpt-builder",
    embedding=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)

# LLM
model = ChatOpenAI()

# RAG chain
chain = (
    RunnableParallel({"context": retriever, "question": RunnablePassthrough()})
    | prompt
    | model
    | StrOutputParser()
)


# Add typing for input
class Question(BaseModel):
    __root__: str


chain = chain.with_types(input_type=Question)