import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
import time

# نموذج NOVA AI
MODEL_NAME = "TheBloke/vicuna-7B-1.1-HF"
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModelForCausalLM.from_pretrained(MODEL_NAME, device_map="auto", torch_dtype=torch.float16)

chat_histories = {}
PERSONALITY = "أنا NOVA AI 😎، كوميدي، مغربي، وودود. نفهم أي حاجة!"

def chat_nova(user_id, message):
    start = time.time()
    if user_id not in chat_histories:
        chat_histories[user_id] = []

    conversation = PERSONALITY + "\n"
    for q, a in chat_histories[user_id]:
        conversation += f"User: {q}\nNOVA AI: {a}\n"
    conversation += f"User: {message}\nNOVA AI:"

    inputs = tokenizer(conversation, return_tensors="pt").to(model.device)
    outputs = model.generate(**inputs, max_new_tokens=200)
    response = tokenizer.decode(outputs[0], skip_special_tokens=True).split("NOVA AI:")[-1].strip()

    chat_histories[user_id].append((message, response))
    if len(chat_histories[user_id]) > 10:
        chat_histories[user_id] = chat_histories[user_id][-10:]

    latency = round(time.time() - start, 2)
    return f"{response}\n\n(⏱ {latency}s)"

# واجهة Gradio Lite
with gr.Blocks() as demo:
    gr.Markdown("## NOVA AI Chat 💡\nشبيهة GPT-5، تجاوب سريع، كوميدية ومغربية.")
    user_id = gr.Textbox(label="ID المستخدم", value="user1")
    message = gr.Textbox(label="أدخل سؤالك")
    output = gr.Textbox(label="رد NOVA AI")
    send_btn = gr.Button("إرسال")
    send_btn.click(chat_nova, inputs=[user_id, message], outputs=output)

demo.launch()