app.py

# -*- coding: utf-8 -*-
"""Untitled68.ipynb

Automatically generated by Colab.

Original file is located at
    https://colab.research.google.com/drive/1h4tpXH6r9B2VZLVwksIkuuVpcrXTUnuJ
"""

import torch
import bitsandbytes as bnb
import transformers
import re
import pandas as pd
import os
import streamlit as st

from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
from langchain.llms import HuggingFacePipeline
from langchain.prompts import PromptTemplate, SystemMessagePromptTemplate, HumanMessagePromptTemplate, ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_community.document_loaders import YoutubeLoader, DataFrameLoader
from langchain_community.vectorstores.utils import filter_complex_metadata
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.vectorstores import FAISS
from langchain.schema.runnable import RunnablePassthrough
from langchain_core.messages import AIMessage, HumanMessage
from langchain_community.llms import HuggingFaceEndpoint
from dotenv import load_dotenv

# Load environment variables from .env file
load_dotenv()

# Get the API token from environment variable
api_token = os.getenv("API_TOKEN")

# Define the repository ID and task
repo_id = "mistralai/Mixtral-8x7B-Instruct-v0.1"
task = "text-generation"

# Initialize the Hugging Face Endpoint
chat_model = HuggingFaceEndpoint(
    huggingfacehub_api_token=api_token,
    repo_id=repo_id,
    task=task
)

template = """
You are a genius trader with extensive knowledge of the financial and stock markets, capable of providing deep and insightful analysis of financial stocks with remarkable accuracy.

**ALWAYS**
Summarize and provide the main insights.
Be as detailed as possible, but don't make up any information that’s not from the context.
If you don't know an answer, say you don't know.
Let's think step by step.

Please ensure responses are informative, accurate, and tailored to the user's queries and preferences.
Use natural language to engage users and provide readable content throughout your response.

{context}
"""

review_system_prompt = SystemMessagePromptTemplate(
    prompt=PromptTemplate(
        input_variables=["context"],
        template=template,
    )
)

review_human_prompt = HumanMessagePromptTemplate(
    prompt=PromptTemplate(
        input_variables=["question"],
        template="{question}",
    )
)
messages = [review_system_prompt, review_human_prompt]

review_prompt_template = ChatPromptTemplate(
    input_variables=["context", "question"],
    messages=messages,
)


def find_youtube_links(text):
    # Define the regular expression pattern for YouTube URLs
    youtube_regex = (r'(https?://(?:www\.)?(?:youtube\.com/watch\?v=|youtu\.be/)[^ \n]+)')
    # Use re.findall() to find all matches in the text
    matches = re.findall(youtube_regex, text)
    return str(' '.join(matches))


# Function to get a response from the model
def get_response(user_query):
    review_chain = (
        {"context": reviews_retriever, "question": RunnablePassthrough()}
        | review_prompt_template
        | chat_model
        | StrOutputParser()
    )
    response = review_chain.invoke(user_query)
    return response

# App config
st.set_page_config(page_title="GOAHEAD.VN", page_icon="🌍")
st.title("Summary and provide insights from youtube news.")

# Initialize session state
if "chat_history" not in st.session_state:
    st.session_state.chat_history = [
        AIMessage(content="Hello, how can I help you?"),
    ]

# Display chat history
for message in st.session_state.chat_history:
    if isinstance(message, AIMessage):
        with st.chat_message("AI"):
            st.write(message.content)
    elif isinstance(message, HumanMessage):
        with st.chat_message("Human"):
            st.write(message.content)

# User input
user_query = st.chat_input("Type your message here...")

if user_query is not None and find_youtube_links(user_query) != "":
    st.session_state.chat_history.append(HumanMessage(content=user_query))

    with st.chat_message("Human"):
        st.markdown(user_query)

    loader = YoutubeLoader.from_youtube_url(
        find_youtube_links(user_query),
        add_video_info=False,
        language=["en", "vi"],
        translation="en",
    )
    docs = loader.load()
    # Convert the loaded documents to a list of dictionaries
    data_list = [
        {
            "source": doc.metadata['source'],
            "page_content": doc.page_content
        }
        for doc in docs
    ]
    df = pd.DataFrame(data_list)
    loader = DataFrameLoader(df, page_content_column='page_content')
    content = loader.load()
    content = filter_complex_metadata(content)

    # Split the document into chunks with a specified chunk size
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1500, chunk_overlap=150)
    all_splits = text_splitter.split_documents(content)

    # Initialize the embedding model
    embedding_model = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L12-v2")

    # Store the document into a vector store with a specific embedding model
    vectorstore = FAISS.from_documents(all_splits, embedding_model)

    reviews_retriever  = vectorstore.as_retriever()

    response = get_response("Help me summary and provide main insights.")

    with st.chat_message("AI"):
        st.write(response)

    st.session_state.chat_history.append(AIMessage(content=response))