app.py

import os
import json
import re
import gradio as gr
import requests
from duckduckgo_search import DDGS
from typing import List
from pydantic import BaseModel, Field
from tempfile import NamedTemporaryFile
from langchain_community.vectorstores import FAISS
from langchain_community.document_loaders import PyPDFLoader
from langchain_community.embeddings import HuggingFaceEmbeddings
from llama_parse import LlamaParse
from langchain_core.documents import Document

# Environment variables and configurations
huggingface_token = os.environ.get("HUGGINGFACE_TOKEN")
llama_cloud_api_key = os.environ.get("LLAMA_CLOUD_API_KEY")

# Initialize LlamaParse
llama_parser = LlamaParse(
    api_key=llama_cloud_api_key,
    result_type="markdown",
    num_workers=4,
    verbose=True,
    language="en",
)

def load_document(file: NamedTemporaryFile, parser: str = "llamaparse") -> List[Document]:
    """Loads and splits the document into pages."""
    if parser == "pypdf":
        loader = PyPDFLoader(file.name)
        return loader.load_and_split()
    elif parser == "llamaparse":
        try:
            documents = llama_parser.load_data(file.name)
            return [Document(page_content=doc.text, metadata={"source": file.name}) for doc in documents]
        except Exception as e:
            print(f"Error using Llama Parse: {str(e)}")
            print("Falling back to PyPDF parser")
            loader = PyPDFLoader(file.name)
            return loader.load_and_split()
    else:
        raise ValueError("Invalid parser specified. Use 'pypdf' or 'llamaparse'.")

def get_embeddings():
    return HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2")

def update_vectors(files, parser):
    if not files:
        return "Please upload at least one PDF file."
    
    embed = get_embeddings()
    total_chunks = 0
    
    all_data = []
    for file in files:
        data = load_document(file, parser)
        all_data.extend(data)
        total_chunks += len(data)
    
    if os.path.exists("faiss_database"):
        database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
        database.add_documents(all_data)
    else:
        database = FAISS.from_documents(all_data, embed)
    
    database.save_local("faiss_database")
    
    return f"Vector store updated successfully. Processed {total_chunks} chunks from {len(files)} files using {parser}."

def generate_chunked_response(prompt, max_tokens=1000, max_chunks=5, temperature=0.3, repetition_penalty=1.1):
    API_URL = "https://api-inference.huggingface.co/models/meta-llama/Meta-Llama-3.1-8B-Instruct"
    headers = {"Authorization": f"Bearer {huggingface_token}"}
    payload = {
        "inputs": prompt,
        "parameters": {
            "max_new_tokens": max_tokens,
            "temperature": temperature,
            "top_p": 0.4,
            "top_k": 40,
            "repetition_penalty": repetition_penalty,
            "stop": ["</s>", "[/INST]"]
        }
    }
    
    full_response = ""
    for _ in range(max_chunks):
        response = requests.post(API_URL, headers=headers, json=payload)
        if response.status_code == 200:
            result = response.json()
            if isinstance(result, list) and len(result) > 0:
                chunk = result[0].get('generated_text', '')
                
                # Remove any part of the chunk that's already in full_response
                new_content = chunk[len(full_response):].strip()
                
                if not new_content:
                    break  # No new content, so we're done
                
                full_response += new_content
                
                if chunk.endswith((".", "!", "?", "</s>", "[/INST]")):
                    break
                
                # Update the prompt for the next iteration
                payload["inputs"] = full_response
            else:
                break
        else:
            break
    
    # Clean up the response
    clean_response = re.sub(r'<s>\[INST\].*?\[/INST\]\s*', '', full_response, flags=re.DOTALL)
    clean_response = clean_response.replace("Using the following context:", "").strip()
    clean_response = clean_response.replace("Using the following context from the PDF documents:", "").strip()
    
    return clean_response

def duckduckgo_search(query):
    with DDGS() as ddgs:
        results = ddgs.text(query, max_results=5)
    return results

class CitingSources(BaseModel):
    sources: List[str] = Field(
        ...,
        description="List of sources to cite. Should be an URL of the source."
    )

def get_response_from_pdf(query, temperature=0.7, repetition_penalty=1.1):
    embed = get_embeddings()
    if os.path.exists("faiss_database"):
        database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
    else:
        return "No documents available. Please upload PDF documents to answer questions."

    retriever = database.as_retriever()
    relevant_docs = retriever.get_relevant_documents(query)
    context_str = "\n".join([doc.page_content for doc in relevant_docs])

    prompt = f"""<s>[INST] Using the following context from the PDF documents:
{context_str}
Write a detailed and complete response that answers the following user question: '{query}'
Do not include a list of sources in your response. [/INST]"""

    generated_text = generate_chunked_response(prompt, temperature=temperature, repetition_penalty=repetition_penalty)

    # Clean the response
    clean_text = re.sub(r'<s>\[INST\].*?\[/INST\]\s*', '', generated_text, flags=re.DOTALL)
    clean_text = clean_text.replace("Using the following context from the PDF documents:", "").strip()

    return clean_text

def get_response_with_search(query, temperature=0.7, repetition_penalty=1.1):
    search_results = duckduckgo_search(query)
    context = "\n".join(f"{result['title']}\n{result['body']}\nSource: {result['href']}\n" 
                        for result in search_results if 'body' in result)
    
    prompt = f"""<s>[INST] Using the following context:
{context}
Write a detailed and complete research document that fulfills the following user request: '{query}'
After writing the document, please provide a list of sources used in your response. [/INST]"""
    
    generated_text = generate_chunked_response(prompt, temperature=temperature, repetition_penalty=repetition_penalty)
    
    # Clean the response
    clean_text = re.sub(r'<s>\[INST\].*?\[/INST\]\s*', '', generated_text, flags=re.DOTALL)
    clean_text = clean_text.replace("Using the following context:", "").strip()
    
    # Split the content and sources
    parts = clean_text.split("Sources:", 1)
    main_content = parts[0].strip()
    sources = parts[1].strip() if len(parts) > 1 else ""
    
    return main_content, sources

def chatbot_interface(message, history, use_web_search, temperature, repetition_penalty):
    if use_web_search:
        main_content, sources = get_response_with_search(message, temperature, repetition_penalty)
        formatted_response = f"{main_content}\n\nSources:\n{sources}"
    else:
        response = get_response_from_pdf(message, temperature, repetition_penalty)
        formatted_response = response

    history.append((message, formatted_response))
    return history

# Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("# AI-powered Web Search and PDF Chat Assistant")
    
    with gr.Row():
        file_input = gr.Files(label="Upload your PDF documents", file_types=[".pdf"])
        parser_dropdown = gr.Dropdown(choices=["pypdf", "llamaparse"], label="Select PDF Parser", value="llamaparse")
        update_button = gr.Button("Upload Document")
    
    update_output = gr.Textbox(label="Update Status")
    update_button.click(update_vectors, inputs=[file_input, parser_dropdown], outputs=update_output)
    
    chatbot = gr.Chatbot(label="Conversation")
    msg = gr.Textbox(label="Ask a question")
    use_web_search = gr.Checkbox(label="Use Web Search", value=False)
    
    with gr.Row():
        temperature_slider = gr.Slider(minimum=0.1, maximum=1.0, value=0.7, step=0.1, label="Temperature")
        repetition_penalty_slider = gr.Slider(minimum=1.0, maximum=2.0, value=1.1, step=0.1, label="Repetition Penalty")
    
    submit = gr.Button("Submit")

    gr.Examples(
        examples=[
            ["What are the latest developments in AI?"],
            ["Tell me about recent updates on GitHub"],
            ["What are the best hotels in Galapagos, Ecuador?"],
            ["Summarize recent advancements in Python programming"],
        ],
        inputs=msg,
    )

    submit.click(chatbot_interface, 
                 inputs=[msg, chatbot, use_web_search, temperature_slider, repetition_penalty_slider], 
                 outputs=[chatbot])
    msg.submit(chatbot_interface, 
               inputs=[msg, chatbot, use_web_search, temperature_slider, repetition_penalty_slider], 
               outputs=[chatbot])

    gr.Markdown(
    """
    ## How to use
    1. Upload PDF documents using the file input at the top.
    2. Select the PDF parser (pypdf or llamaparse) and click "Upload Document" to update the vector store.
    3. Ask questions in the textbox. 
    4. Toggle "Use Web Search" to switch between PDF chat and web search.
    5. Adjust Temperature and Repetition Penalty sliders to fine-tune the response generation.
    6. Click "Submit" or press Enter to get a response.
    """
    )

if __name__ == "__main__":
    demo.launch(share=True)