app.py

import gradio as gr
from huggingface_hub import InferenceClient
import requests
from deep_translator import GoogleTranslator

client = InferenceClient("mistralai/Mistral-Nemo-Instruct-2407")

def translate_to_english(text: str) -> str:
    try:
        return GoogleTranslator(source='auto', target='en').translate(text)
    except Exception:
        return text

def translate_to_bisaya(text: str) -> str:
    try:
        return GoogleTranslator(source='auto', target='ceb').translate(text)
    except Exception:
        return text

def get_internet_data(query: str) -> str:
    """
    Uses Qwant's free search API to fetch a snippet based on the query.
    """
    url = "https://api.qwant.com/v3/search/web"
    params = {
        "q": query,
        "count": 10,
        "offset": 0,
        "t": "web",
        "safesearch": 1,
        "locale": "en_US",
        "uiv": 4,
    }
    try:
        response = requests.get(url, params=params, timeout=5)
        response.raise_for_status()
        data = response.json()
        items = data.get("data", {}).get("result", {}).get("items", [])
        if items:
            snippet = items[0].get("desc", "")
            if not snippet:
                snippet = items[0].get("title", "")
        else:
            snippet = "Wala koy nakuha nga impormasyon gikan sa Qwant search."
    except Exception:
        snippet = "Naay problema sa pagkuha sa impormasyon gikan sa Qwant search."
    return snippet

def respond(message, history: list[tuple[str, str]]):
    # Step 1: Translate the query from Bisaya to English.
    english_query = translate_to_english(message)
    
    # Step 2: Search the web using Qwant's API with the translated query.
    search_result = get_internet_data(english_query)
    
    # Step 3: Translate the search result back to Bisaya.
    bisaya_search_result = translate_to_bisaya(search_result)
    
    # Enrich the original query with the translated search result.
    enriched_message = (
        f"{message}\n\nMga resulta gikan sa internet (isinalin sa bisaya): {bisaya_search_result}"
    )
    
    system_message = (
        "Ikaw usa ka buotan nga Chatbot. Tubaga lang sa binisaya. "
        "Gamiton ang bag-ong kasayuran nga nakuha gikan sa internet. "
        "Ayaw og gamit ug English nga pinulungan."
    )
    max_tokens = 4096
    temperature = 0.6
    top_p = 0.95

    messages = [{"role": "system", "content": system_message}]
    for user_text, assistant_text in history:
        if user_text:
            messages.append({"role": "user", "content": user_text})
        if assistant_text:
            messages.append({"role": "assistant", "content": assistant_text})
    messages.append({"role": "user", "content": enriched_message})
    
    # Get the complete response from the model.
    full_response = ""
    for token_message in client.chat_completion(
        messages,
        max_tokens=max_tokens,
        stream=True,
        temperature=temperature,
        top_p=top_p,
    ):
        token = token_message.choices[0].delta.get("content", "")
        if not token:
            break
        full_response += token
        if len(full_response) > 3000:
            break

    # Translate the final response to Bisaya.
    final_response = translate_to_bisaya(full_response)
    yield final_response

demo = gr.ChatInterface(respond)

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