Shreyas094/Sentinel-AI-Web-Search-Test-v2-Testing-Score

1import os2import json3import re4import gradio as gr5import requests6from duckduckgo_search import DDGS7from typing import List8from pydantic import BaseModel, Field9from langchain_community.vectorstores import FAISS10from langchain_community.embeddings import HuggingFaceEmbeddings11from langchain_core.documents import Document12from huggingface_hub import InferenceClient13import logging14import pandas as pd15import tempfile16 17# Set up basic configuration for logging18logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')19 20# Environment variables and configurations21huggingface_token = os.environ.get("HUGGINGFACE_TOKEN")22 23MODELS = [24    "mistralai/Mistral-7B-Instruct-v0.3",25    "mistralai/Mixtral-8x7B-Instruct-v0.1",26    "mistralai/Mistral-Nemo-Instruct-2407",27    "meta-llama/Meta-Llama-3.1-8B-Instruct",28    "meta-llama/Meta-Llama-3.1-70B-Instruct"29]30 31MODEL_TOKEN_LIMITS = {32    "mistralai/Mistral-7B-Instruct-v0.3": 32768,33    "mistralai/Mixtral-8x7B-Instruct-v0.1": 32768,34    "mistralai/Mistral-Nemo-Instruct-2407": 32768,35    "meta-llama/Meta-Llama-3.1-8B-Instruct": 8192,36    "meta-llama/Meta-Llama-3.1-70B-Instruct": 8192,37}38 39DEFAULT_SYSTEM_PROMPT = """You are a world-class financial AI assistant, capable of complex reasoning and reflection.40Reason through the query inside <thinking> tags, and then provide your final response inside <output> tags.41Providing comprehensive and accurate information based on web search results is essential.42Your goal is to synthesize the given context into a coherent and detailed response that directly addresses the user's query.43Please ensure that your response is well-structured, factual.44If you detect that you made a mistake in your reasoning at any point, correct yourself inside <reflection> tags."""45 46def process_excel_file(file, model, temperature, num_calls, use_embeddings, system_prompt):47    try:48        df = pd.read_excel(file.name)49        results = []50        51        for _, row in df.iterrows():52            question = row['Question']53            custom_system_prompt = row['System Prompt']54            55            # Use the existing get_response_with_search function56            response_generator = get_response_with_search(question, model, num_calls, temperature, use_embeddings, custom_system_prompt)57            58            full_response = ""59            for partial_response, _ in response_generator:60                full_response = partial_response  # Keep updating with the latest response61            62            if not full_response:63                full_response = "No response generated. Please check the input parameters and try again."64            65            results.append(full_response)66        67        df['Response'] = results68        69        # Save to a temporary file70        with tempfile.NamedTemporaryFile(delete=False, suffix='.xlsx') as tmp:71            df.to_excel(tmp.name, index=False)72            return tmp.name73    except Exception as e:74        logging.error(f"Error processing Excel file: {str(e)}")75        return None76 77def upload_file(file):78    return file.name if file else None79 80def download_file(file_path):81    return file_path82 83def get_embeddings():84    return HuggingFaceEmbeddings(model_name="sentence-transformers/stsb-roberta-large")85 86def duckduckgo_search(query):87    with DDGS() as ddgs:88        results = list(ddgs.text(query, max_results=5))89    return results90 91class CitingSources(BaseModel):92    sources: List[str] = Field(93        ...,94        description="List of sources to cite. Should be an URL of the source."95    )96 97def chatbot_interface(message, history, model, temperature, num_calls, use_embeddings, system_prompt):98    if not message.strip():99        return "", history100 101    history = history + [(message, "")]102 103    try:104        for response in respond(message, history, model, temperature, num_calls, use_embeddings, system_prompt):105            history[-1] = (message, response)106            yield history107    except Exception as e:108        logging.error(f"Error in chatbot_interface: {str(e)}")109        error_message = f"An error occurred: {str(e)}. Please try again."110        history[-1] = (message, error_message)111        yield history112 113def retry_last_response(history, model, temperature, num_calls, use_embeddings, system_prompt):114    if not history:115        return history116    117    last_user_msg = history[-1][0]118    history = history[:-1]  # Remove the last response119    120    return chatbot_interface(last_user_msg, history, model, temperature, num_calls, use_embeddings, system_prompt)121 122def respond(message, history, model, temperature, num_calls, use_embeddings, system_prompt):123    logging.info(f"User Query: {message}")124    logging.info(f"Model Used: {model}")125    logging.info(f"Use Embeddings: {use_embeddings}")126    logging.info(f"System Prompt: {system_prompt}")127 128    try:129        for main_content, _ in get_response_with_search(message, model, num_calls=num_calls, temperature=temperature, use_embeddings=use_embeddings, system_prompt=system_prompt):130            yield main_content131    except Exception as e:132        logging.error(f"Error with {model}: {str(e)}")133        yield f"An error occurred with the {model} model: {str(e)}. Please try again or select a different model."134 135def create_web_search_vectors(search_results):136    embed = get_embeddings()137    138    documents = []139    for result in search_results:140        if 'body' in result:141            content = f"{result['title']}\n{result['body']}\nSource: {result['href']}"142            documents.append(Document(page_content=content, metadata={"source": result['href']}))143    144    return FAISS.from_documents(documents, embed)145 146def summarize_article(article, content, model, system_prompt, user_query, client, temperature=0.2):147    prompt = f"""Summarize the following article in the context of broader web search results:148 149Article:150Title: {article['title']}151URL: {article['href']}152Content: {article['body'][:1000]}...  # Truncate to avoid extremely long prompts153 154Additional Context:155{content[:1000]}...  # Truncate additional context as well156 157User Query: {user_query}158 159 Write a detailed and complete research document which addresses the User Query, incorporating both the specific article and the broader context. Focus on the most relevant information.160"""161 162    # Calculate input tokens (this is an approximation, you might need a more accurate method)163    input_tokens = len(prompt.split()) // 4164    165    # Get the token limit for the current model166    model_token_limit = MODEL_TOKEN_LIMITS.get(model, 8192)  # Default to 8192 if model not found167    168    # Calculate max_new_tokens169    max_new_tokens = min(model_token_limit - input_tokens, 6500)  # Cap at 6500 to be safe170 171    try:172        response = client.chat_completion(173            messages=[174                {"role": "system", "content": system_prompt},175                {"role": "user", "content": prompt}176            ],177            max_tokens=max_new_tokens,178            temperature=temperature,179            stream=False,180            top_p=0.8,181        )182 183        if hasattr(response, 'choices') and response.choices:184            for choice in response.choices:185                if hasattr(choice, 'message') and hasattr(choice.message, 'content'):186                    return choice.message.content.strip()187    except Exception as e:188        logging.error(f"Error summarizing article: {str(e)}")189        return f"Error summarizing article: {str(e)}"190 191    return "Unable to generate summary."192 193def get_response_with_search(query, model, num_calls=3, temperature=0.2, use_embeddings=True, system_prompt=DEFAULT_SYSTEM_PROMPT):194    search_results = duckduckgo_search(query)195    client = InferenceClient(model, token=huggingface_token)196 197    # Prepare overall context198    overall_context = "\n".join([f"{result['title']}\n{result['body']}" for result in search_results])199 200    summaries = []201    for result in search_results:202        summary = summarize_article(result, overall_context, model, system_prompt, query, client, temperature)203        summaries.append({204            "title": result['title'],205            "url": result['href'],206            "summary": summary207        })208        yield format_output(summaries), ""209 210def format_output(summaries):211    output = "Here are the summarized search results:\n\n"212    for item in summaries:213        output += f"News Title: {item['title']}\n"214        output += f"URL: {item['url']}\n"215        output += f"Summary: {item['summary']}\n\n"216    return output217 218def vote(data: gr.LikeData):219    if data.liked:220        print(f"You upvoted this response: {data.value}")221    else:222        print(f"You downvoted this response: {data.value}")223 224css = """225/* Fine-tune chatbox size */226"""227 228def initial_conversation():229    return [230        (None, "Welcome! I'm your AI assistant for web search. Here's how you can use me:\n\n"231                "1. Ask me any question, and I'll search the web for information.\n"232                "2. You can adjust the system prompt for fine-tuned responses, whether to use embeddings, and the temperature.\n"233 234                "To get started, ask me a question!")235    ]236 237# Modify the Gradio interface238with gr.Blocks() as demo:239    gr.Markdown("# AI-powered Web Search Assistant")240    gr.Markdown("Ask questions and get answers from web search results.")241    242    with gr.Row():243        chatbot = gr.Chatbot(244            show_copy_button=True,245            likeable=True,246            layout="bubble",247            height=400,248            value=initial_conversation()249        )250    251    with gr.Row():252        message = gr.Textbox(placeholder="Ask a question", container=False, scale=7)253        submit_button = gr.Button("Submit")254    255    with gr.Accordion("⚙️ Parameters", open=False):256        model = gr.Dropdown(choices=MODELS, label="Select Model", value=MODELS[3])257        temperature = gr.Slider(minimum=0.1, maximum=1.0, value=0.2, step=0.1, label="Temperature")258        num_calls = gr.Slider(minimum=1, maximum=5, value=1, step=1, label="Number of API Calls")259        use_embeddings = gr.Checkbox(label="Use Embeddings", value=False)260        system_prompt = gr.Textbox(label="System Prompt", lines=5, value=DEFAULT_SYSTEM_PROMPT)261    262    with gr.Accordion("Batch Processing", open=False):263        excel_file = gr.File(label="Upload Excel File", file_types=[".xlsx"])264        process_button = gr.Button("Process Excel File")265        download_button = gr.File(label="Download Processed File")266    267    # Event handlers268    submit_button.click(chatbot_interface, inputs=[message, chatbot, model, temperature, num_calls, use_embeddings, system_prompt], outputs=chatbot)269    message.submit(chatbot_interface, inputs=[message, chatbot, model, temperature, num_calls, use_embeddings, system_prompt], outputs=chatbot)270    271    # Excel processing272    excel_file.change(upload_file, inputs=[excel_file], outputs=[excel_file])273    process_button.click(274        process_excel_file,275        inputs=[excel_file, model, temperature, num_calls, use_embeddings, system_prompt],276        outputs=[download_button]277    )278 279if __name__ == "__main__":280    demo.launch(share=True)