hitsense/sense_faq

1import pandas as pd2import os3import streamlit as st4from langchain.embeddings.openai import OpenAIEmbeddings5from langchain.text_splitter import CharacterTextSplitter6from langchain.docstore.document import Document7from langchain.vectorstores import FAISS8from langchain.chains.question_answering import load_qa_chain9from langchain.llms import OpenAI, HuggingFaceHub10from langchain.chat_models import ChatOpenAI11from langchain import PromptTemplate, LLMChain12from langchain.embeddings import HuggingFaceHubEmbeddings13 14 15# hf_llm = HuggingFaceHub(16#     repo_id="google/flan-t5-xl",17#     # repo_id="tloen/alpaca-lora-7b",18#     huggingfacehub_api_token = os.environ['HUGGINGFACEHUB_API_TOKEN'],19#     model_kwargs={"temperature":0.001, "max_new_tokens":500}20#                        ) 21    22# @st.cache_data23# def text_processing(text):24#     text_splitter = CharacterTextSplitter(chunk_size=500, chunk_overlap=0)25#     docs = text_splitter.create_documents([text])26#     faiss_index = FAISS.load_local("sense_faq_handbook_faiss_index", OpenAIEmbeddings())27#     # faiss_index = FAISS.from_documents(docs,28#     #                                    HuggingFaceHubEmbeddings(),29#     #                                    # OpenAIEmbeddings()30#     #                                   )31#     return faiss_index32 33@st.cache_data34def build_chain():35  return load_qa_chain(36      # hf_llm,37      ChatOpenAI(temperature=0, model_name="gpt-3.5-turbo"), 38      chain_type="stuff"39  )40 41# @st.cache_data42def run_chain(query, chain):43    index = FAISS.load_local("sense_faq_handbook_faiss_index", OpenAIEmbeddings())44    docs = index.similarity_search(query)45    return chain.run(input_documents=docs, question=query)46