app.py

import os
import gradio as gr
from pathlib import Path
from llama_cpp import Llama
import requests

MODEL_URL = "https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.1-GGUF/resolve/main/mistral-7b-instruct-v0.1.Q4_0.gguf"
MODEL_PATH = "mistral-7b.Q4_0.gguf"

# 🔁 Download if needed
def download_model():
    if not Path(MODEL_PATH).exists():
        print("⬇️ Downloading model...")
        with requests.get(MODEL_URL, stream=True) as r:
            with open(MODEL_PATH, "wb") as f:
                for chunk in r.iter_content(chunk_size=8192):
                    f.write(chunk)
        print("✅ Model ready.")

download_model()

# 🚀 Load LLM
llm = Llama(
    model_path=MODEL_PATH,
    n_ctx=4096,
    n_batch=128,
    n_threads=os.cpu_count(),
    use_mlock=True,
    use_mmap=True,
    chat_format="mistral-instruct"
)

# 💬 Smart prompt formatting
def format_chat(chat_history):
    messages = []
    for q, a in chat_history:
        messages.append({"role": "user", "content": q})
        messages.append({"role": "assistant", "content": a})
    return messages

# 🔄 Main logic
def respond(message, chat_history):
    messages = format_chat(chat_history)
    messages.append({"role": "user", "content": message})

    output = llm.create_chat_completion(
        messages,
        max_tokens=300,
        temperature=0.7,
        stop=["</s>", "<|endoftext|>"]
    )

    response = output["choices"][0]["message"]["content"]
    chat_history.append((message, response))
    return "", chat_history

# 🖼️ Custom UI
with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown("<h1 style='text-align: center;'>🤖 Mistral-7B-Instruct Chat (Aria CPU)</h1>")
    chatbot = gr.Chatbot(show_label=False, avatar_images=("🤖", "🧑‍💻"))
    msg = gr.Textbox(placeholder="Ask something smart...", label="You")
    clear = gr.Button("🧹 Clear")

    state = gr.State([])

    msg.submit(respond, [msg, state], [msg, chatbot, state])
    clear.click(lambda: ([], []), None, [chatbot, state])

demo.launch()