custom_summarization_app.py

import openai
import streamlit as st
import os
from langchain.document_loaders import PyPDFLoader
from langchain import PromptTemplate
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.chains.summarize import load_summarize_chain
from langchain.chat_models import ChatOpenAI

openai.api_key = os.environ["OPENAI_API_KEY"]

@st.cache_data
def setup_documents(pdf_file_path,chunk_size,chunk_overlap):
    loader = PyPDFLoader(pdf_file_path)
    docs_raw = loader.load()
    docs_raw_text = [doc.page_content for doc in docs_raw]
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=chunk_size,
                                                   chunk_overlap=chunk_overlap)
    docs = text_splitter.create_documents(docs_raw_text)
    
    return docs
    


def custom_summary(docs, llm, custom_prompt, chain_type, num_summaries):
    custom_prompt = custom_prompt + """:\n {text}"""
    COMBINE_PROMPT = PromptTemplate(template=custom_prompt, input_variables = ["text"])
    MAP_PROMPT = PromptTemplate(template="Summarize:\n{text}", input_variables=["text"])
    if chain_type == "map_reduce":
        chain = load_summarize_chain(llm,chain_type=chain_type,
                                     map_prompt=MAP_PROMPT,
                                     combine_prompt=COMBINE_PROMPT)
    else:
        chain = load_summarize_chain(llm,chain_type=chain_type)
    
    summaries = []
    for i in range(num_summaries):
        summary_output = chain({"input_documents": docs}, return_only_outputs=True)["output_text"]
        summaries.append(summary_output)
    
    return summaries

@st.cache_data
def color_chunks(text: str, chunk_size: int, overlap_size: int) -> str:
    overlap_color = "#808080"
    chunk_colors = ["#a8d08d", "#c6dbef", "#e6550d", "#fd8d3c", "#fdae6b", "#fdd0a2"] # Different shades of green for chunks
    
    colored_text = ""
    overlap = ""
    color_index = 0
    for i in range(0, len(text), chunk_size-overlap_size):
        chunk = text[i:i+chunk_size]
        if overlap:
            colored_text += f'<mark style="background-color: {overlap_color};">{overlap}</mark>'
        chunk = chunk[len(overlap):]
        colored_text += f'<mark style="background-color: {chunk_colors[color_index]};">{chunk}</mark>'
        color_index = (color_index + 1) % len(chunk_colors)
        overlap = text[i+chunk_size-overlap_size:i+chunk_size]

    return colored_text


def main():
    st.set_page_config(layout="wide")
    st.title("Custom Summarization App")
    llm = st.sidebar.selectbox("LLM",["ChatGPT", "GPT4", "Other (open source in the future)"])
    chain_type = st.sidebar.selectbox("Chain Type", ["map_reduce", "stuff", "refine"])
    chunk_size = st.sidebar.slider("Chunk Size", min_value=20, max_value = 10000,
                                   step=10, value=2000)
    chunk_overlap = st.sidebar.slider("Chunk Overlap", min_value=5, max_value = 5000,
                                   step=10, value=200)
    
    if st.sidebar.checkbox("Debug chunk size"):
        st.header("Interactive Text Chunk Visualization")

        text_input = st.text_area("Input Text", "This is a test text to showcase the functionality of the interactive text chunk visualizer.")

        # Set the minimum to 1, the maximum to 5000 and default to 100
        html_code = color_chunks(text_input, chunk_size, chunk_overlap)
        st.markdown(html_code, unsafe_allow_html=True)
    
    else:
        user_prompt = st.text_input("Enter the custom summary prompt")
        pdf_file_path = st.text_input("Enther the pdf file path")
        
        temperature = st.sidebar.number_input("Set the ChatGPT Temperature",
                                              min_value = 0.0,
                                              max_value=1.0,
                                              step=0.1,
                                              value=0.5)
        num_summaries = st.sidebar.number_input("Number of summaries",
                                                min_value = 1, 
                                                max_value = 10,
                                                step = 1,
                                                value=1)
        if pdf_file_path != "":
            docs = setup_documents(pdf_file_path, chunk_size, chunk_overlap)
            st.write("PDF loaded successfully")
        
            if llm=="ChatGPT":
                llm = ChatOpenAI(temperature=temperature)
            elif llm=="GPT4":
                llm = ChatOpenAI(model_name="gpt-4",temperature=temperature)
            else:
                st.write("Using ChatGPT while open source models are not implemented!")
                llm = ChatOpenAI(temperature=temperature)
            
            if st.button("Summarize"):
                result = custom_summary(docs, llm, user_prompt, chain_type, num_summaries)
                st.write("Summary:")
                for summary in result:
                    st.write(summary)
            
    
if __name__=="__main__":
    main()