app.py

import tempfile
import os
import streamlit as st

from llama_index.llms.gemini import Gemini
from llama_index.llms.huggingface import HuggingFaceLLM
from llama_index.llms.mistralai import MistralAI
from llama_index.llms.openai import OpenAI
from llama_index.core import (
    VectorStoreIndex,
    Settings,
)

from llama_parse import LlamaParse

from streamlit_pdf_viewer import pdf_viewer

# Global configurations
from llama_index.core import set_global_handler
set_global_handler("langfuse")
st.set_page_config(layout="wide")

with st.sidebar:
    st.title('Document Summarization and QA System')
    # st.markdown('''
    # ## About this application
    # Upload a pdf to ask questions about it. This retrieval-augmented generation (RAG) workflow uses:
    # - [Streamlit](https://streamlit.io/)
    # - [LlamaIndex](https://docs.llamaindex.ai/en/stable/)
    # - [OpenAI](https://platform.openai.com/docs/models)
    # ''')

    # st.write('Made by ***Nate Mahynski***')
    # st.write('nathan.mahynski@nist.gov')

    # Select Provider
    provider = st.selectbox(
        label="Select LLM Provider",
        options=['google', 'huggingface', 'mistralai', 'openai'],
        index=0
    )

    # Select LLM
    if provider == 'google':
        llm_list = ['gemini']
    elif provider == 'huggingface':
        llm_list = []
    elif provider == 'mistralai':
        llm_list =[]
    elif provider == 'openai':
        llm_list = ['gpt-3.5-turbo', 'gpt-4', 'gpt-4-turbo', 'gpt-4o']
    else:
        llm_list = []

    llm_name = st.selectbox(
        label="Select LLM Model",
        options=llm_list,
        index=0
    )

    # Temperature
    temperature = st.slider(
        "Temperature",
        min_value=0.0, 
        max_value=1.0, 
        value=0.0, 
        step=0.05, 
    )

    max_output_tokens = 4096

    # Create LLM
    if provider == 'openai':
        llm = OpenAI(
            model=llm_name, 
            temperature=temperature,
            max_tokens=max_tokens
        )
        # Global tokenization needs to be consistent with LLM
        # https://docs.llamaindex.ai/en/stable/module_guides/models/llms/
        Settings.tokenizer = tiktoken.encoding_for_model(llm_name).encode
        Settings.num_output = max_tokens
        Settings.context_window = 4096 # max possible
    

    # Enter LLM Token
    llm_token = st.text_input(
        "Enter your LLM token",
        value=None
    )

    # Enter parsing Token
    parse_token = st.text_input(
        "Enter your LlamaParse token",
        value=None
    )

    uploaded_file = st.file_uploader(
        "Choose a PDF file to upload", 
        type=['pdf'], 
        accept_multiple_files=False
    )

    if uploaded_file is not None:
        # Parse the file
        temp_dir = tempfile.TemporaryDirectory()
        parser = LlamaParse(
            api_key=parse_token,  # can also be set in your env as LLAMA_CLOUD_API_KEY
            result_type="text"  # "markdown" and "text" are available
        )

        filename = os.path.join('./', uploaded_file.name)
        with open(filename, "wb") as f:
            f.write(uploaded_file.getvalue())

        parsed_document = parser.load_data(filename)
        temp_dir.cleanup()

col1, col2 = st.columns(2)

with col1:
    st.write(uploaded_file)
    st.write(parsed_document)

with col2:
    if uploaded_file is not None:
        # Display the pdf
        bytes_data = uploaded_file.getvalue()
        pdf_viewer(input=bytes_data, width=700)