import gradio as gr
from huggingface_hub import InferenceClient
import json
from bs4 import BeautifulSoup
import requests
from transformers import LlavaProcessor, LlavaForConditionalGeneration, TextIteratorStreamer
from threading import Thread
import torch

model_id = "llava-hf/llava-interleave-qwen-0.5b-hf"
processor = LlavaProcessor.from_pretrained(model_id)
model = LlavaForConditionalGeneration.from_pretrained(model_id)
model.to("cpu")

def extract_text_from_webpage(html_content):
    soup = BeautifulSoup(html_content, 'html.parser')
    for tag in soup(["script", "style", "header", "footer"]):
        tag.extract()
    return soup.get_text(strip=True)

def search(query):
    term = query
    all_results = []
    max_chars_per_page = 8000
    with requests.Session() as session:
        resp = session.get(
            url="https://www.google.com/search",
            headers={"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:109.0) Gecko/20100101 Firefox/111.0"},
            params={"q": term, "num": 3, "udm": 14},
            timeout=5,
        )
        resp.raise_for_status()
        soup = BeautifulSoup(resp.text, "html.parser")
        result_block = soup.find_all("div", attrs={"class": "g"})
        for result in result_block:
            link = result.find("a", href=True)["href"]
            try:
                webpage = session.get(link, headers={"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:109.0) Gecko/20100101 Firefox/111.0"}, timeout=5)
                webpage.raise_for_status()
                visible_text = extract_text_from_webpage(webpage.text)
                if len(visible_text) > max_chars_per_page:
                    visible_text = visible_text[:max_chars_per_page]
                all_results.append({"link": link, "text": visible_text})
            except requests.exceptions.RequestException:
                all_results.append({"link": link, "text": None})
    return all_results

# Initialize inference clients for different models
client_gemma = InferenceClient("mistralai/Mistral-7B-Instruct-v0.3")
client_mixtral = InferenceClient("NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO")
client_llama = InferenceClient("meta-llama/Meta-Llama-3-8B-Instruct")

# Define the main chat function
def respond(question, history):
    func_caller = []

    user_prompt = question
    functions_metadata = [
        {"type": "function", "function": {"name": "web_search", "description": "Search query on google", "parameters": {"type": "object", "properties": {"query": {"type": "string", "description": "web search query"}}, "required": ["query"]}}},
        {"type": "function", "function": {"name": "general_query", "description": "Reply general query of USER", "parameters": {"type": "object", "properties": {"prompt": {"type": "string", "description": "A detailed prompt"}}, "required": ["prompt"]}}},
    ]

    for msg in history:
        func_caller.append({"role": "user", "content": f"{str(msg[0])}"})
        func_caller.append({"role": "assistant", "content": f"{str(msg[1])}"})

    func_caller.append({"role": "user", "content": f'[SYSTEM]You are a helpful assistant. You have access to the following functions: \n {str(functions_metadata)}\n\nTo use these functions respond with:\n<functioncall> {{ "name": "function_name", "arguments": {{ "arg_1": "value_1", "arg_1": "value_1", ... }} }}  </functioncall>  [USER] {question}'})
    
    response = client_gemma.chat_completion(func_caller, max_tokens=200)
    response = str(response)
    try:
        response = response[int(response.find("{")):int(response.rindex("</"))]
    except:
        response = response[int(response.find("{")):(int(response.rfind("}"))+1)]
    response = response.replace("\\n", "")
    response = response.replace("\\'", "'")
    response = response.replace('\\"', '"')
    response = response.replace('\\', '')
    print(f"\n{response}")
    
    try:
        json_data = json.loads(str(response))
        if json_data["name"] == "web_search":
            query = json_data["arguments"]["query"]
            web_results = search(query)
            web2 = ' '.join([f"Link: {res['link']}\nText: {res['text']}\n\n" for res in web_results])
            messages = f"system\nYou are OpenCHAT mini a helpful assistant. You are provided with WEB results from which you can find informations to answer users query in Structured and More better way. You do not say Unnecesarry things Only say thing which is important and relevant. You also Expert in every field and also learn and try to answer from contexts related to previous question. Try your best to give best response possible to user. You also try to show emotions using Emojis and reply like human, use short forms, friendly tone and emotions."
            for msg in history:
                messages += f"\nuser\n{str(msg[0])}"
                messages += f"\nassistant\n{str(msg[1])}"
            messages += f"\nuser\n{question}\nweb_result\n{web2}\nassistant\n"
            stream = client_mixtral.text_generation(messages, max_new_tokens=2000, do_sample=True, stream=True, details=True, return_full_text=False)
            output = ""
            for response in stream:
                if not response.token.text == "":
                    output += response.token.text
                    yield output
        else:
            messages = f"system\nYou are OpenCHAT mini a helpful assistant. You answers users query like human friend. You are also Expert in every field and also learn and try to answer from contexts related to previous question. Try your best to give best response possible to user. You also try to show emotions using Emojis and reply like human, use short forms, friendly tone and emotions."
            for msg in history:
                messages += f"\nuser\n{str(msg[0])}"
                messages += f"\nassistant\n{str(msg[1])}"
            messages += f"\nuser\n{question}\nassistant\n"
            stream = client_llama.text_generation(messages, max_new_tokens=2000, do_sample=True, stream=True, details=True, return_full_text=False)
            output = ""
            for response in stream:
                if not response.token.text == "":
                    output += response.token.text
                    yield output
    except:
        messages = f"system\nYou are OpenCHAT mini a helpful assistant. You answers users query like human friend. You are also Expert in every field and also learn and try to answer from contexts related to previous question. Try your best to give best response possible to user. You also try to show emotions using Emojis and reply like human, use short forms, friendly tone and emotions."
        for msg in history:
            messages += f"\nuser\n{str(msg[0])}"
            messages += f"\nassistant\n{str(msg[1])}"
        messages += f"\nuser\n{question}\nassistant\n"
        stream = client_llama.text_generation(messages, max_new_tokens=2000, do_sample=True, stream=True, details=True, return_full_text=False)
        output = ""
        for response in stream:
            if not response.token.text == "":
                output += response.token.text
                yield output

# Create the Gradio interface
demo = gr.Interface(
    fn=respond,
    inputs=gr.Textbox(label="Question"),
    outputs=gr.Textbox(label="Response"),
    description="# OpenGPT 4o mini\n### You can engage in chat, generate images, perform web searches, and Q&A with images.",
    examples=[
        {"question": "Hi, who are you?"},
        {"question": "What's the current price of Bitcoin?"},
        {"question": "Search and tell me what's the release date of llama 3 400b."},
        {"question": "Write me a Python function to calculate the first 10 digits of the Fibonacci sequence."},
    ],
)
demo.launch(show_error=True)