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 # Example for document loading (from url), splitting, and creating vectostore """ # Load from langchain_community.document_loaders import WebBaseLoader loader = WebBaseLoader("https://lilianweng.github.io/posts/2023-06-23-agent/") data = loader.load() # Split from langchain_text_splitters import RecursiveCharacterTextSplitter text_splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=0) all_splits = text_splitter.split_documents(data) # Add to vectorDB vectorstore = Chroma.from_documents(documents=all_splits, collection_name="rag-chroma", embedding=OpenAIEmbeddings(), ) retriever = vectorstore.as_retriever() """ # Embed a single document as a test vectorstore = Chroma.from_texts( ["harrison worked at kensho"], collection_name="rag-chroma", 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)