streamlit_app.py

import streamlit as st
from streamlit_feedback import streamlit_feedback

import os
import pandas as pd
import base64
from io import BytesIO
import sqlite3
import uuid
import yaml

import chromadb
from llama_index.core import (
            VectorStoreIndex, 
            SimpleDirectoryReader,
            StorageContext,
            Document
)
from llama_index.vector_stores.chroma.base import ChromaVectorStore
from llama_index.embeddings.huggingface.base import HuggingFaceEmbedding
from llama_index.llms.openai import OpenAI
from llama_index.core.memory import ChatMemoryBuffer
from llama_index.core.tools import QueryEngineTool
from llama_index.agent.openai import OpenAIAgent
from llama_index.core import Settings

from vision_api import get_transcribed_text
from qna_prompting import get_qna_question_tool, evaluate_qna_answer_tool
from prompt_engineering import (
            system_content, 
            textbook_content, 
            winnie_the_pooh_prompt, 
            introduction_line
)

import nest_asyncio
nest_asyncio.apply()

# App title
st.set_page_config(page_title="🐻📚 Study Bear 🍯")
openai_api = os.getenv("OPENAI_API_KEY")

with open("./config/model_config_advanced.yml", "r") as file_reader:
    model_config = yaml.safe_load(file_reader)

input_files = model_config["input_data"]["source"]
embedding_model = model_config["embeddings"]["embedding_base_model"]
fine_tuned_path = model_config["embeddings"]["fine_tuned_embedding_model"]
persisted_vector_db = model_config["vector_store"]["persisted_path"]
questionaire_db_path = model_config["questionaire_data"]["db_path"]

data_df = pd.DataFrame(
    {
        "Completion": [30, 40, 100, 10],
    }
)
data_df.index = ["Chapter 1", "Chapter 2", "Chapter 3", "Chapter 4"]

bear_img_path = "./resource/disney-cuties-little-winnie-the-pooh-emoticon.png"
piglet_img_path = "./resource/disney-cuties-piglet-emoticon.png"

# Replicate Credentials
with st.sidebar:
    st.title("🍯🐝 Study Bear 🐻💭")
    st.write("Just like Pooh needs honey, success requires hard work – no shortcuts allowed!")
    wtp_mode = st.toggle('Winnie-the-Pooh mode', value=False)
    if wtp_mode:
        system_content = system_content + winnie_the_pooh_prompt
    textbook_content = system_content + textbook_content

    if openai_api:
        pass
    elif "OPENAI_API_KEY" in st.secrets:
        st.success("API key already provided!", icon="✅")
        openai_api = st.secrets["OPENAI_API_KEY"]
    else:
        openai_api = st.text_input("Enter OpenAI API token:", type="password")
        if not (openai_api.startswith("sk-") and len(openai_api)==51):
            st.warning("Please enter your credentials!", icon="⚠️")
        else:
            st.success("Proceed to entering your prompt message!", icon="👉")

    ### for streamlit purpose
    os.environ["OPENAI_API_KEY"] = openai_api

    st.subheader("Models and parameters")
    selected_model = st.sidebar.selectbox(label="Choose an OpenAI model", 
                                          options=["gpt-3.5-turbo-0125", "gpt-4-0125-preview"], 
                                          index=1,
                                          key="selected_model")
    temperature = st.sidebar.slider("temperature", min_value=0.0, max_value=2.0, 
                                    value=0.0, step=0.01)
    st.data_editor(
        data_df,
        column_config={
            "Completion": st.column_config.ProgressColumn(
                            "Completion %",
                            help="Percentage of content covered",
                            format="%.1f%%",
                            min_value=0,
                            max_value=100,
            ),
        },
        hide_index=False,
    )

    st.markdown("📖 Reach out to SakiMilo to learn how to create this app!")

if "init" not in st.session_state.keys():
    st.session_state.init = {"warm_started": "No"}
    st.session_state.feedback = False

if "image_prompt" not in st.session_state.keys():
    st.session_state.image_prompt = False

# Store LLM generated responses
if "messages" not in st.session_state.keys():
    st.session_state.messages = [{"role": "assistant", 
                                  "content": introduction_line,
                                  "type": "text"}]

if "feedback_key" not in st.session_state:
    st.session_state.feedback_key = 0

if "release_file" not in st.session_state:
    st.session_state.release_file = "false"

if "question_id" not in st.session_state:
    st.session_state.question_id = None

if "qna_answer" not in st.session_state:
    st.session_state.qna_answer = None

if "reasons" not in st.session_state:
    st.session_state.reasons = None

if "user_id" not in st.session_state:
    st.session_state.user_id = str(uuid.uuid4())

def clear_chat_history():

    st.session_state.messages = [{"role": "assistant", 
                                  "content": introduction_line,
                                  "type": "text"}]
    chat_engine = get_query_engine(input_files=input_files, 
                                   llm_model=selected_model, 
                                   temperature=temperature,
                                   embedding_model=embedding_model,
                                   fine_tuned_path=fine_tuned_path,
                                   system_content=system_content,
                                   persisted_vector_db=persisted_vector_db)
    chat_engine.reset()
    st.toast("yumyum, what was I saying again? 🐻💬", icon="🍯")

def clear_question_history(user_id):

    con = sqlite3.connect(questionaire_db_path)
    cur = con.cursor()
    sql_string = f"""
                  DELETE FROM answer_tbl
                  WHERE user_id='{user_id}'
    """
    res = cur.execute(sql_string)
    con.commit()
    con.close()
    st.toast("the tale of one thousand and one questions, reset! 🧨🧨", icon="📝")

st.sidebar.button("Clear Chat History", on_click=clear_chat_history)
st.sidebar.button("Clear Question History", 
                  on_click=clear_question_history,
                  kwargs={"user_id": st.session_state.user_id})
if st.sidebar.button("I want to submit a feedback!"):
    st.session_state.feedback = True
    st.session_state.feedback_key += 1  # overwrite feedback component

@st.cache_resource
def get_document_object(input_files):
    documents = SimpleDirectoryReader(input_files=input_files).load_data()
    document = Document(text="\n\n".join([doc.text for doc in documents]))
    return document

@st.cache_resource
def get_llm_object(selected_model, temperature):
    llm = OpenAI(model=selected_model, temperature=temperature)
    return llm

@st.cache_resource
def get_embedding_model(model_name, fine_tuned_path=None):
    if fine_tuned_path is None:
        print(f"loading from `{model_name}` from huggingface")
        embed_model = HuggingFaceEmbedding(model_name=model_name)
    else:
        print(f"loading from local `{fine_tuned_path}`")
        embed_model = fine_tuned_path
    return embed_model

@st.cache_resource
def get_query_engine(input_files, llm_model, temperature,
                     embedding_model, fine_tuned_path,
                     system_content, persisted_vector_db):
    
    llm = get_llm_object(llm_model, temperature)
    embedded_model = get_embedding_model(
                        model_name=embedding_model, 
                        fine_tuned_path=fine_tuned_path
    )
    Settings.llm = llm
    Settings.chunk_size = 1024
    Settings.embed_model = embedded_model

    if os.path.exists(persisted_vector_db):
        print("loading from vector database - chroma")
        db = chromadb.PersistentClient(path=persisted_vector_db)
        chroma_collection = db.get_or_create_collection("quickstart")
        vector_store = ChromaVectorStore(chroma_collection=chroma_collection)
        storage_context = StorageContext.from_defaults(vector_store=vector_store)

        index = VectorStoreIndex.from_vector_store(
            vector_store=vector_store,
            storage_context=storage_context
        )
    else:
        print("create new chroma vector database..")
        documents = SimpleDirectoryReader(input_files=input_files).load_data()
        
        db = chromadb.PersistentClient(path=persisted_vector_db)
        chroma_collection = db.get_or_create_collection("quickstart")
        vector_store = ChromaVectorStore(chroma_collection=chroma_collection)
        
        nodes = Settings.node_parser.get_nodes_from_documents(documents)
        storage_context = StorageContext.from_defaults(vector_store=vector_store)
        storage_context.docstore.add_documents(nodes)

        index = VectorStoreIndex(nodes, storage_context=storage_context)
    
    memory = ChatMemoryBuffer.from_defaults(token_limit=100_000)
    hi_content_engine = index.as_query_engine(
                            memory=memory,
                            system_prompt=system_content,
                            similarity_top_k=10,
                            verbose=True,
                            streaming=True
    )
    hi_textbook_query_description = """
        Use this tool to extract content from the query engine,
        which is built by ingesting textbook content from `Health Insurance 7th Edition`,
        that has 15 chapters in total. When user wants to learn more about a 
        particular chapter, this tool will help to assist user to get better
        understanding of the content of the textbook. 
    """
    
    hi_query_tool = QueryEngineTool.from_defaults(
                        query_engine=hi_content_engine,
                        name="health_insurance_textbook_query_engine",
                        description=hi_textbook_query_description
    )

    agent = OpenAIAgent.from_tools(tools=[
                                        hi_query_tool, 
                                        get_qna_question_tool,
                                        evaluate_qna_answer_tool
                                    ],
                                   max_function_calls=1,
                                   llm=llm, 
                                   verbose=True,
                                   system_prompt=textbook_content)
    print("loaded AI agent, let's begin the chat!")
    print("="*50)
    print("")

    return agent

def generate_llm_response(prompt_input, tool_choice="auto"):
    chat_agent = get_query_engine(input_files=input_files, 
                                   llm_model=selected_model, 
                                   temperature=temperature,
                                   embedding_model=embedding_model,
                                   fine_tuned_path=fine_tuned_path,
                                   system_content=system_content,
                                   persisted_vector_db=persisted_vector_db)
    
    # st.session_state.messages
    response = chat_agent.stream_chat(prompt_input, tool_choice=tool_choice)
    return response

def handle_feedback(user_response):
    st.toast("✔️ Feedback received!")
    st.session_state.feedback = False

def handle_image_upload():
    st.session_state.release_file = "true"

# Warm start
if st.session_state.init["warm_started"] == "No":
    clear_chat_history()
    st.session_state.init["warm_started"] = "Yes"

# Image upload option
with st.sidebar:
    image_file = st.file_uploader("Upload your image here...", 
                                  type=["png", "jpeg", "jpg"],
                                  on_change=handle_image_upload)

    if st.session_state.release_file == "true" and image_file:
        with st.spinner("Uploading..."):
            b64string = base64.b64encode(image_file.read()).decode('utf-8')
            message = {
                    "role": "user", 
                    "content": b64string,
                    "type": "image"}
            st.session_state.messages.append(message)

            transcribed_msg = get_transcribed_text(b64string)
            message = {
                    "role": "admin", 
                    "content": transcribed_msg,
                    "type": "text"}
            st.session_state.messages.append(message)
            st.session_state.release_file = "false"

# Display or clear chat messages
for message in st.session_state.messages:
    if message["role"] == "admin":
        continue
    elif message["role"] == "user":
        avatar = piglet_img_path
    elif message["role"] == "assistant":
        avatar = bear_img_path
    
    with st.chat_message(message["role"], avatar=avatar):
        if message["type"] == "text":
            st.write(message["content"])
        elif message["type"] == "image":
            img_io = BytesIO(base64.b64decode(message["content"].encode("utf-8")))
            st.image(img_io)

# User-provided prompt
if prompt := st.chat_input(disabled=not openai_api):
    st.session_state.messages.append({"role": "user", 
                                      "content": prompt, 
                                      "type": "text"})
    with st.chat_message("user", avatar=piglet_img_path):
        st.write(prompt)

# Retrieve text prompt from image submission
if prompt is None and \
   st.session_state.messages[-1]["role"] == "admin":
    st.session_state.image_prompt = True
    prompt = st.session_state.messages[-1]["content"]

# Generate a new response if last message is not from assistant
if st.session_state.messages[-1]["role"] != "assistant":
    with st.chat_message("assistant", avatar=bear_img_path):
        with st.spinner("🧸💤 Thinking... 🐻💭"):
            if st.session_state.image_prompt:
                response = generate_llm_response(
                                prompt, 
                                tool_choice="health_insurance_textbook_query_engine"
                            )
                st.session_state.image_prompt = False
            else:
                response = generate_llm_response(prompt, tool_choice="auto")
            placeholder = st.empty()
            full_response = ""
            for token in response.response_gen:
                token = token.replace("\n", "  \n") \
                             .replace("$", "\$") \
                             .replace("\[", "$$")
                full_response += token
                placeholder.markdown(full_response)
            placeholder.markdown(full_response)

    message = {"role": "assistant", 
               "content": full_response,
               "type": "text"}
    st.session_state.messages.append(message)

# Trigger feedback
if st.session_state.feedback:
    result = streamlit_feedback(
                feedback_type="thumbs",
                optional_text_label="[Optional] Please provide an explanation",
                on_submit=handle_feedback,
                key=f"feedback_{st.session_state.feedback_key}"
    )