Spaces:
Paused
Paused
import os | |
from typing import List | |
from langchain.embeddings.openai import OpenAIEmbeddings | |
from langchain.vectorstores.pinecone import Pinecone | |
from langchain.chains import ConversationalRetrievalChain | |
from langchain.chat_models import ChatOpenAI | |
from langchain.memory import ChatMessageHistory, ConversationBufferMemory | |
from langchain_core.prompts import PromptTemplate | |
from langchain.docstore.document import Document | |
import pinecone | |
import chainlit as cl | |
from cleanlab_studio import Studio | |
pinecone.init( | |
api_key=os.environ.get("PINECONE_API_KEY"), | |
environment=os.environ.get("PINECONE_ENV"), | |
) | |
studio = Studio(os.getenv("CLEANLAB_API_KEY")) | |
tlm = studio.TLM(quality_preset='high') | |
index_name = "tracker" | |
embeddings = OpenAIEmbeddings() | |
welcome_message = "Welcome to the Transparency Tracker! Ask me any question related to Anti-Corruption." | |
async def start(): | |
await cl.Message(content=welcome_message,disable_human_feedback=True).send() | |
docsearch = Pinecone.from_existing_index( | |
index_name=index_name, embedding=embeddings | |
) | |
message_history = ChatMessageHistory() | |
memory = ConversationBufferMemory( | |
memory_key="chat_history", | |
output_key="answer", | |
chat_memory=message_history, | |
return_messages=True, | |
) | |
with open('./prompt.txt','r') as f: | |
template = f.read() | |
prompt = PromptTemplate(input_variables=["context", "question"],template=template) | |
chain = ConversationalRetrievalChain.from_llm( | |
llm = ChatOpenAI( | |
model_name="gpt-3.5-turbo", | |
temperature=0, | |
streaming=True), | |
chain_type="stuff", | |
retriever=docsearch.as_retriever(search_kwargs={'k': 3}), # I only want maximum of three document back with the highest similarity score | |
memory=memory, | |
return_source_documents=True, | |
combine_docs_chain_kwargs={"prompt": prompt} | |
) | |
cl.user_session.set("chain", chain) | |
async def evaluate_response(action): | |
await action.remove() | |
arr = action.value.split('|||') | |
confidence_score = tlm.get_confidence_score(arr[0], response=arr[1]) | |
await cl.Message(content=f"Confidence Score: {confidence_score}",disable_human_feedback=True).send() | |
async def main(message: cl.Message): | |
chain = cl.user_session.get("chain") | |
cb = cl.AsyncLangchainCallbackHandler() | |
res = await chain.acall(message.content, callbacks=[cb]) | |
answer = res["answer"] | |
source_documents = res["source_documents"] | |
text_elements = [] | |
if source_documents: | |
for source_idx, source_doc in enumerate(source_documents): | |
source_name = f"source_{source_idx}" | |
text_elements.append( | |
cl.Text(content=source_doc.page_content, name=source_name) | |
) | |
source_names = [text_el.name for text_el in text_elements] | |
if source_names: | |
answer += f"\nSources: {', '.join(source_names)}" | |
else: | |
answer += "\nNo sources found" | |
actions = [ | |
cl.Action(name="eval_button",value=f"{message.content}|||{answer}",label='Evaluate with CleanLab',description="Evaluate with CleanLab TLM (*may take a moment*)") | |
] | |
await cl.Message(content=answer, elements=text_elements, actions=actions).send() | |