src/app.py

import os
from threading import Lock
from typing import Any, Dict, Optional, Tuple

import gradio as gr
from langchain.chains import ConversationalRetrievalChain
from langchain.chat_models import ChatOpenAI
from langchain.memory import ConversationBufferMemory
from langchain.prompts.chat import (ChatPromptTemplate,
                                    HumanMessagePromptTemplate,
                                    SystemMessagePromptTemplate)

from src.core.chunking import chunk_file
from src.core.embedding import embed_files
from src.core.parsing import read_file

VECTOR_STORE = "faiss"
MODEL = "openai"
EMBEDDING = "openai"
MODEL = "gpt-3.5-turbo-16k"
K = 5
USE_VERBOSE = True
API_KEY = os.environ["OPENAI_API_KEY"]
system_template = """
The context below contains excerpts from 'Ben Hogan's Five Lessions'. You must only use the information in the context below to formulate your response. If there is not enough information to formulate a response, you must respond with
"I'm sorry, but I can't find the answer to your question in, Ben Hogan's Five Lessons."

Here is the context:
{context}
{chat_history}
"""

# Create the chat prompt templates
messages = [
  SystemMessagePromptTemplate.from_template(system_template),
  HumanMessagePromptTemplate.from_template("{question}")
]
qa_prompt = ChatPromptTemplate.from_messages(messages)

class AnswerConversationBufferMemory(ConversationBufferMemory):
  def save_context(self, inputs: Dict[str, Any], outputs: Dict[str, str]) -> None:
    return super(AnswerConversationBufferMemory, self).save_context(inputs,{'response': outputs['answer']})

def getretriever():
  with open("./resources/Ben_Hogans_Five_Lessons.pdf", 'rb') as uploaded_file:
    try:
      file = read_file(uploaded_file)
    except Exception as e:
      print(e)

  chunked_file = chunk_file(file, chunk_size=512, chunk_overlap=0)
  folder_index = embed_files(
    files=[chunked_file],
    embedding=EMBEDDING,
    vector_store=VECTOR_STORE,
    openai_api_key=API_KEY,
  )
  return folder_index.index.as_retriever(verbose=True, search_type="similarity", search_kwargs={"k": K})

retriever = getretriever()

def getanswer(chain, question, history):
  if hasattr(chain, "value"):
    chain = chain.value
  if hasattr(history, "value"):
    history = history.value
  if hasattr(question, "value"):
    question = question.value

  history = history or []
  lock = Lock()
  lock.acquire()
  try:
    output = chain({"question": question})
    output = output["answer"]
    history.append((question, output))
  except Exception as e:
    raise e
  finally:
    lock.release()
  return history, history, gr.update(value="")

def load_chain(inputs = None):
  llm = ChatOpenAI(
        openai_api_key=API_KEY,
        model_name=MODEL,
        verbose=True)
  chain = ConversationalRetrievalChain.from_llm(
    llm,
    retriever=retriever,
    return_source_documents=USE_VERBOSE,
    memory=AnswerConversationBufferMemory(memory_key="chat_history", return_messages=True),
    verbose=USE_VERBOSE,
    combine_docs_chain_kwargs={"prompt": qa_prompt})
  return chain

with gr.Blocks() as block:
  with gr.Row():
    with gr.Column(scale=0.75):
      with gr.Row():
        gr.Markdown("<h1>Ben Hogan's Five Lessons</h1>")
      with gr.Row():
        gr.Markdown("by Ben Hogan")
      chatbot = gr.Chatbot(elem_id="chatbot").style(height=600)

      with gr.Row():
          message = gr.Textbox(
              label="",
              placeholder="Ask Ben...",
              lines=1,
          )
      with gr.Row():
          submit = gr.Button(value="Send", variant="primary", scale=1)

      state = gr.State()
      chain_state = gr.State(load_chain)

      submit.click(getanswer, inputs=[chain_state, message, state], outputs=[chatbot, state, message])
      message.submit(getanswer, inputs=[chain_state, message, state], outputs=[chatbot, state, message])

    with gr.Column(scale=0.25):
      with gr.Row():
        gr.Markdown("<h1><center>Suggestions</center></h1>")
      ex1 = gr.Button(value="What is this book about?", variant="primary")
      ex1.click(getanswer, inputs=[chain_state, ex1, state], outputs=[chatbot, state, message])
      ex2 = gr.Button(value="What are the core fundamentals Ben teaches?", variant="primary")
      ex2.click(getanswer, inputs=[chain_state, ex2, state], outputs=[chatbot, state, message])
      ex3 = gr.Button(value="How can I improve my swing?", variant="primary")
      ex3.click(getanswer, inputs=[chain_state, ex3, state], outputs=[chatbot, state, message])

block.launch(debug=True)