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
from huggingface_hub import InferenceClient
import inspect
import logging


# Set up basic configuration for logging
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')

# Environment variables and configurations
huggingface_token = os.environ.get("HUGGINGFACE_TOKEN")
llama_cloud_api_key = os.environ.get("LLAMA_CLOUD_API_KEY")
ACCOUNT_ID = os.environ.get("CLOUDFARE_ACCOUNT_ID")
API_TOKEN = os.environ.get("CLOUDFLARE_AUTH_TOKEN")
API_BASE_URL = "https://api.cloudflare.com/client/v4/accounts/a17f03e0f049ccae0c15cdcf3b9737ce/ai/run/"

print(f"ACCOUNT_ID: {ACCOUNT_ID}")
print(f"CLOUDFLARE_AUTH_TOKEN: {API_TOKEN[:5]}..." if API_TOKEN else "Not set")

MODELS = [
    "mistralai/Mistral-7B-Instruct-v0.3",
    "mistralai/Mixtral-8x7B-Instruct-v0.1",
    "@cf/meta/llama-3.1-8b-instruct",
    "mistralai/Mistral-Nemo-Instruct-2407"
]

# 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/stsb-roberta-large")

def update_vectors(files, parser):
    global uploaded_documents
    logging.info(f"Entering update_vectors with {len(files)} files and parser: {parser}")
    
    if not files:
        logging.warning("No files provided for update_vectors")
        return "Please upload at least one PDF file.", gr.CheckboxGroup(
            choices=[doc["name"] for doc in uploaded_documents],
            value=[doc["name"] for doc in uploaded_documents if doc["selected"]],
            label="Select documents to query"
        )
    
    embed = get_embeddings()
    total_chunks = 0
    
    all_data = []
    for file in files:
        logging.info(f"Processing file: {file.name}")
        try:
            data = load_document(file, parser)
            logging.info(f"Loaded {len(data)} chunks from {file.name}")
            all_data.extend(data)
            total_chunks += len(data)
            # Append new documents instead of replacing
            if not any(doc["name"] == file.name for doc in uploaded_documents):
                uploaded_documents.append({"name": file.name, "selected": True})
                logging.info(f"Added new document to uploaded_documents: {file.name}")
            else:
                logging.info(f"Document already exists in uploaded_documents: {file.name}")
        except Exception as e:
            logging.error(f"Error processing file {file.name}: {str(e)}")
    
    logging.info(f"Total chunks processed: {total_chunks}")
    
    if os.path.exists("faiss_database"):
        logging.info("Updating existing FAISS database")
        database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
        database.add_documents(all_data)
    else:
        logging.info("Creating new FAISS database")
        database = FAISS.from_documents(all_data, embed)
    
    database.save_local("faiss_database")
    logging.info("FAISS database saved")
    
    return f"Vector store updated successfully. Processed {total_chunks} chunks from {len(files)} files using {parser}.", gr.CheckboxGroup(
        choices=[doc["name"] for doc in uploaded_documents],
        value=[doc["name"] for doc in uploaded_documents if doc["selected"]],
        label="Select documents to query"
    )

def generate_chunked_response(prompt, model, max_tokens=10000, num_calls=3, temperature=0.2, should_stop=False):
    print(f"Starting generate_chunked_response with {num_calls} calls")
    full_response = ""
    messages = [{"role": "user", "content": prompt}]
    
    if model == "@cf/meta/llama-3.1-8b-instruct":
        # Cloudflare API
        for i in range(num_calls):
            print(f"Starting Cloudflare API call {i+1}")
            if should_stop:
                print("Stop clicked, breaking loop")
                break
            try:
                response = requests.post(
                    f"https://api.cloudflare.com/client/v4/accounts/{ACCOUNT_ID}/ai/run/@cf/meta/llama-3.1-8b-instruct",
                    headers={"Authorization": f"Bearer {API_TOKEN}"},
                    json={
                        "stream": true,
                        "messages": [
                            {"role": "system", "content": "You are a friendly assistant"},
                            {"role": "user", "content": prompt}
                        ],
                        "max_tokens": max_tokens,
                        "temperature": temperature
                    },
                    stream=true
                )
                
                for line in response.iter_lines():
                    if should_stop:
                        print("Stop clicked during streaming, breaking")
                        break
                    if line:
                        try:
                            json_data = json.loads(line.decode('utf-8').split('data: ')[1])
                            chunk = json_data['response']
                            full_response += chunk
                        except json.JSONDecodeError:
                            continue
                print(f"Cloudflare API call {i+1} completed")
            except Exception as e:
                print(f"Error in generating response from Cloudflare: {str(e)}")
    else:
        # Original Hugging Face API logic
        client = InferenceClient(model, token=huggingface_token)
        
        for i in range(num_calls):
            print(f"Starting Hugging Face API call {i+1}")
            if should_stop:
                print("Stop clicked, breaking loop")
                break
            try:
                for message in client.chat_completion(
                    messages=messages,
                    max_tokens=max_tokens,
                    temperature=temperature,
                    stream=True,
                ):
                    if should_stop:
                        print("Stop clicked during streaming, breaking")
                        break
                    if message.choices and message.choices[0].delta and message.choices[0].delta.content:
                        chunk = message.choices[0].delta.content
                        full_response += chunk
                print(f"Hugging Face API call {i+1} completed")
            except Exception as e:
                print(f"Error in generating response from Hugging Face: {str(e)}")
    
    # 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()
    
    # Remove duplicate paragraphs and sentences
    paragraphs = clean_response.split('\n\n')
    unique_paragraphs = []
    for paragraph in paragraphs:
        if paragraph not in unique_paragraphs:
            sentences = paragraph.split('. ')
            unique_sentences = []
            for sentence in sentences:
                if sentence not in unique_sentences:
                    unique_sentences.append(sentence)
            unique_paragraphs.append('. '.join(unique_sentences))
    
    final_response = '\n\n'.join(unique_paragraphs)
    
    print(f"Final clean response: {final_response[:100]}...")
    return final_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 chatbot_interface(message, history, use_web_search, model, temperature, num_calls):
    if not message.strip():
        return "", history

    history = history + [(message, "")]

    try:
        for response in respond(message, history, model, temperature, num_calls, use_web_search):
            history[-1] = (message, response)
            yield history
    except gr.CancelledError:
        yield history
    except Exception as e:
        logging.error(f"Unexpected error in chatbot_interface: {str(e)}")
        history[-1] = (message, f"An unexpected error occurred: {str(e)}")
        yield history

def retry_last_response(history, use_web_search, model, temperature, num_calls):
    if not history:
        return history
    
    last_user_msg = history[-1][0]
    history = history[:-1]  # Remove the last response
    
    return chatbot_interface(last_user_msg, history, use_web_search, model, temperature, num_calls)

def respond(message, history, model, temperature, num_calls, use_web_search, selected_docs, instruction_key):
    logging.info(f"User Query: {message}")
    logging.info(f"Model Used: {model}")
    logging.info(f"Search Type: {'Web Search' if use_web_search else 'PDF Search'}")
    logging.info(f"Selected Documents: {selected_docs}")
    logging.info(f"Instruction Key: {instruction_key}")

    try:
        if instruction_key and instruction_key != "None":
            # This is a summary generation request
            instruction = INSTRUCTION_PROMPTS[instruction_key]
            context_str = get_context_for_summary(selected_docs)
            message = f"{instruction}\n\nUsing the following context from the PDF documents:\n{context_str}\nGenerate a detailed summary."
            use_web_search = False  # Ensure we use PDF search for summaries

        if use_web_search:
            for main_content, sources in get_response_with_search(message, model, num_calls=num_calls, temperature=temperature):
                response = f"{main_content}\n\n{sources}"
                first_line = response.split('\n')[0] if response else ''
                logging.info(f"Generated Response (first line): {first_line}")
                yield response
        else:
            embed = get_embeddings()
            if os.path.exists("faiss_database"):
                database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
                retriever = database.as_retriever()
                
                # Filter relevant documents based on user selection
                all_relevant_docs = retriever.get_relevant_documents(message)
                relevant_docs = [doc for doc in all_relevant_docs if doc.metadata["source"] in selected_docs]
                
                if not relevant_docs:
                    yield "No relevant information found in the selected documents. Please try selecting different documents or rephrasing your query."
                    return

                context_str = "\n".join([doc.page_content for doc in relevant_docs])
            else:
                context_str = "No documents available."
                yield "No documents available. Please upload PDF documents to answer questions."
                return
            
            if model == "@cf/meta/llama-3.1-8b-instruct":
                # Use Cloudflare API
                for partial_response in get_response_from_cloudflare(prompt="", context=context_str, query=message, num_calls=num_calls, temperature=temperature, search_type="pdf"):
                    first_line = partial_response.split('\n')[0] if partial_response else ''
                    logging.info(f"Generated Response (first line): {first_line}")
                    yield partial_response
            else:
                # Use Hugging Face API
                for partial_response in get_response_from_pdf(message, model, selected_docs, num_calls=num_calls, temperature=temperature):
                    first_line = partial_response.split('\n')[0] if partial_response else ''
                    logging.info(f"Generated Response (first line): {first_line}")
                    yield partial_response

    except Exception as e:
        logging.error(f"Error with {model}: {str(e)}")
        if "microsoft/Phi-3-mini-4k-instruct" in model:
            logging.info("Falling back to Mistral model due to Phi-3 error")
            fallback_model = "mistralai/Mistral-7B-Instruct-v0.3"
            yield from respond(message, history, fallback_model, temperature, num_calls, use_web_search, selected_docs, instruction_key)
        else:
            yield f"An error occurred with the {model} model: {str(e)}. Please try again or select a different model."

logging.basicConfig(level=logging.DEBUG)

INSTRUCTION_PROMPTS = {
    "Asset Managers": "Summarize the key financial metrics, assets under management, and performance highlights for this asset management company.",
    "Consumer Finance Companies": "Provide a summary of the company's loan portfolio, interest income, credit quality, and key operational metrics.",
    "Mortgage REITs": "Summarize the REIT's mortgage-backed securities portfolio, net interest income, book value per share, and dividend yield.",
    # Add more instruction prompts as needed
}

def get_response_from_cloudflare(prompt, context, query, num_calls=3, temperature=0.2, search_type="pdf"):
    headers = {
        "Authorization": f"Bearer {API_TOKEN}",
        "Content-Type": "application/json"
    }
    model = "@cf/meta/llama-3.1-8b-instruct"

    if search_type == "pdf":
        instruction = f"""Using the following context from the PDF documents:
{context}
Write a detailed and complete response that answers the following user question: '{query}'"""
    else:  # web search
        instruction = f"""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."""

    inputs = [
        {"role": "system", "content": instruction},
        {"role": "user", "content": query}
    ]

    payload = {
        "messages": inputs,
        "stream": True,
        "temperature": temperature,
        "max_tokens": 32000
    }

    full_response = ""
    for i in range(num_calls):
        try:
            with requests.post(f"{API_BASE_URL}{model}", headers=headers, json=payload, stream=True) as response:
                if response.status_code == 200:
                    for line in response.iter_lines():
                        if line:
                            try:
                                json_response = json.loads(line.decode('utf-8').split('data: ')[1])
                                if 'response' in json_response:
                                    chunk = json_response['response']
                                    full_response += chunk
                                    yield full_response
                            except (json.JSONDecodeError, IndexError) as e:
                                logging.error(f"Error parsing streaming response: {str(e)}")
                                continue
                else:
                    logging.error(f"HTTP Error: {response.status_code}, Response: {response.text}")
                    yield f"I apologize, but I encountered an HTTP error: {response.status_code}. Please try again later."
        except Exception as e:
            logging.error(f"Error in generating response from Cloudflare: {str(e)}")
            yield f"I apologize, but an error occurred: {str(e)}. Please try again later."
    
    if not full_response:
        yield "I apologize, but I couldn't generate a response at this time. Please try again later."

def get_response_with_search(query, model, num_calls=3, temperature=0.2):
    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"""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."""

    if model == "@cf/meta/llama-3.1-8b-instruct":
        # Use Cloudflare API
        for response in get_response_from_cloudflare(prompt="", context=context, query=query, num_calls=num_calls, temperature=temperature, search_type="web"):
            yield response, ""  # Yield streaming response without sources
    else:
        # Use Hugging Face API
        client = InferenceClient(model, token=huggingface_token)
        
        main_content = ""
        for i in range(num_calls):
            for message in client.chat_completion(
                messages=[{"role": "user", "content": prompt}],
                max_tokens=10000,
                temperature=temperature,
                stream=True,
            ):
                if message.choices and message.choices[0].delta and message.choices[0].delta.content:
                    chunk = message.choices[0].delta.content
                    main_content += chunk
                    yield main_content, ""  # Yield partial main content without sources

def get_response_from_pdf(query, model, selected_docs, num_calls=3, temperature=0.2):
    logging.info(f"Entering get_response_from_pdf with query: {query}, model: {model}, selected_docs: {selected_docs}")
    
    embed = get_embeddings()
    if os.path.exists("faiss_database"):
        logging.info("Loading FAISS database")
        database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
    else:
        logging.warning("No FAISS database found")
        yield "No documents available. Please upload PDF documents to answer questions."
        return

    retriever = database.as_retriever()
    logging.info(f"Retrieving relevant documents for query: {query}")
    relevant_docs = retriever.get_relevant_documents(query)
    logging.info(f"Number of relevant documents retrieved: {len(relevant_docs)}")
    
    # Filter relevant_docs based on selected documents
    filtered_docs = [doc for doc in relevant_docs if doc.metadata["source"] in selected_docs]
    logging.info(f"Number of filtered documents: {len(filtered_docs)}")
    
    if not filtered_docs:
        logging.warning(f"No relevant information found in the selected documents: {selected_docs}")
        yield "No relevant information found in the selected documents. Please try selecting different documents or rephrasing your query."
        return

    for doc in filtered_docs:
        logging.info(f"Document source: {doc.metadata['source']}")
        logging.info(f"Document content preview: {doc.page_content[:100]}...")  # Log first 100 characters of each document

    context_str = "\n".join([doc.page_content for doc in filtered_docs])
    logging.info(f"Total context length: {len(context_str)}")

    if model == "@cf/meta/llama-3.1-8b-instruct":
        logging.info("Using Cloudflare API")
        # Use Cloudflare API with the retrieved context
        for response in get_response_from_cloudflare(prompt="", context=context_str, query=query, num_calls=num_calls, temperature=temperature, search_type="pdf"):
            yield response
    else:
        logging.info("Using Hugging Face API")
        # Use Hugging Face API
        prompt = f"""Using the following context from the PDF documents:
{context_str}
Write a detailed and complete response that answers the following user question: '{query}'"""
        
        client = InferenceClient(model, token=huggingface_token)
        
        response = ""
        for i in range(num_calls):
            logging.info(f"API call {i+1}/{num_calls}")
            for message in client.chat_completion(
                messages=[{"role": "user", "content": prompt}],
                max_tokens=10000,
                temperature=temperature,
                stream=True,
            ):
                if message.choices and message.choices[0].delta and message.choices[0].delta.content:
                    chunk = message.choices[0].delta.content
                    response += chunk
                    yield response  # Yield partial response
        
        logging.info("Finished generating response")

def vote(data: gr.LikeData):
    if data.liked:
        print(f"You upvoted this response: {data.value}")
    else:
        print(f"You downvoted this response: {data.value}")

css = """
/* Fine-tune chatbox size */
.chatbot-container {
    height: 600px !important;
    width: 100% !important;
}
.chatbot-container > div {
    height: 100%;
    width: 100%;
}
"""

def get_context_for_summary(selected_docs):
    embed = get_embeddings()
    if os.path.exists("faiss_database"):
        database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
        retriever = database.as_retriever(search_kwargs={"k": 5})  # Retrieve top 5 most relevant chunks
        
        # Create a generic query that covers common financial summary topics
        generic_query = "financial performance revenue profit assets liabilities cash flow key metrics highlights"
        
        relevant_docs = retriever.get_relevant_documents(generic_query)
        filtered_docs = [doc for doc in relevant_docs if doc.metadata["source"] in selected_docs]
        
        if not filtered_docs:
            return "No relevant information found in the selected documents for summary generation."
        
        context_str = "\n".join([doc.page_content for doc in filtered_docs])
        return context_str
    else:
        return "No documents available for summary generation."

def get_context_for_query(query, selected_docs):
    embed = get_embeddings()
    if os.path.exists("faiss_database"):
        database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
        retriever = database.as_retriever(search_kwargs={"k": 3})  # Retrieve top 3 most relevant chunks
        
        relevant_docs = retriever.get_relevant_documents(query)
        filtered_docs = [doc for doc in relevant_docs if doc.metadata["source"] in selected_docs]
        
        if not filtered_docs:
            return "No relevant information found in the selected documents for the given query."
        
        context_str = "\n".join([doc.page_content for doc in filtered_docs])
        return context_str
    else:
        return "No documents available to answer the query."

uploaded_documents = []

def display_documents():
    return gr.CheckboxGroup(
        choices=[doc["name"] for doc in uploaded_documents],
        value=[doc["name"] for doc in uploaded_documents if doc["selected"]],
        label="Select documents to query"
    )

def initial_conversation():
    return [
        (None, "Welcome! I'm your AI assistant for web search and PDF analysis. Here's how you can use me:\n\n"
                "1. Set the toggle for Web Search and PDF Search from the checkbox in Additional Inputs drop down window\n"
                "2. Use web search to find information\n"
                "3. Ask questions about uploaded PDF documents\n"
                "4. Generate summaries for specific entity types\n\n"
                "To get started, upload some PDFs or ask me a question!")
    ]

# Define the checkbox outside the demo block
document_selector = gr.CheckboxGroup(label="Select documents to query")

use_web_search = gr.Checkbox(label="Use Web Search", value=True)

custom_placeholder = "Ask a question (Note: You can toggle between Web Search and PDF Chat in Additional Inputs below)"

instruction_choices = ["None"] + list(INSTRUCTION_PROMPTS.keys())

demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Dropdown(choices=MODELS, label="Select Model", value=MODELS[3]),
        gr.Slider(minimum=0.1, maximum=1.0, value=0.2, step=0.1, label="Temperature"),
        gr.Slider(minimum=1, maximum=5, value=1, step=1, label="Number of API Calls"),
        use_web_search,
        document_selector,
        gr.Dropdown(choices=instruction_choices, label="Select Entity Type for Summary", value="None")
    ],
    title="AI-powered Web Search and PDF Chat Assistant",
    description="Chat with your PDFs, use web search to answer questions, or generate summaries. Select an Entity Type for Summary to generate a specific summary.",
    theme=gr.themes.Soft(
        primary_hue="orange",
        secondary_hue="amber",
        neutral_hue="gray",
        font=[gr.themes.GoogleFont("Exo"), "ui-sans-serif", "system-ui", "sans-serif"]
    ).set(
        body_background_fill_dark="#0c0505",
        block_background_fill_dark="#0c0505",
        block_border_width="1px",
        block_title_background_fill_dark="#1b0f0f",
        input_background_fill_dark="#140b0b",
        button_secondary_background_fill_dark="#140b0b",
        border_color_accent_dark="#1b0f0f",
        border_color_primary_dark="#1b0f0f",
        background_fill_secondary_dark="#0c0505",
        color_accent_soft_dark="transparent",
        code_background_fill_dark="#140b0b"
    ),
    css=css,
    examples=[
        ["Tell me about the contents of the uploaded PDFs."],
        ["What are the main topics discussed in the documents?"],
        ["Can you summarize the key points from the PDFs?"]
    ],
    cache_examples=False,
    analytics_enabled=False,
    textbox=gr.Textbox(placeholder=custom_placeholder, container=False, scale=7),
    chatbot = gr.Chatbot(  
        show_copy_button=True,
        likeable=True,
        layout="bubble",
        height=400,
    )
)

# Add file upload functionality
with demo:
    gr.Markdown("## Upload PDF Documents")

    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 both the output text and the document selector
    update_button.click(update_vectors, 
                        inputs=[file_input, parser_dropdown], 
                        outputs=[update_output, document_selector])

    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. Select the documents you want to query using the checkboxes.
    4. Ask questions in the chat interface. 
    5. Toggle "Use Web Search" to switch between PDF chat and web search.
    6. Adjust Temperature and Number of API Calls to fine-tune the response generation.
    7. Use the provided examples or ask your own questions.
    """
    )

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