import gradio as gr
from llama_cpp import Llama

# --- Load model from Hugging Face Hub (Q8 GGUF) ---
llm = Llama.from_pretrained(
    repo_id="owlninjam/nytheria-3b", 
    filename="gguf_models/nytheria-3b-q8_0.gguf",
    n_ctx=2048,
    n_threads=8  # CPU threads; free tier usually gives 2-4 cores
)

# --- Chat function ---
def chat(message, history):
    if history is None:
        history = []
    response = llm(message, max_tokens=512, stop=["</s>"])
    output = response["choices"][0]["text"].strip()
    history.append(("🧑 You: " + message, "🤖 Nytheria: " + output))
    return history, ""

# --- Build Sexy Chat UI ---
with gr.Blocks(css="""
    #chatbot {height: 600px; overflow: auto;}
    .message.user {background: #4f46e5; color: white; border-radius: 20px; padding: 10px 15px;}
    .message.bot {background: #f3f4f6; color: black; border-radius: 20px; padding: 10px 15px;}
    .wrap {display: flex; gap: 10px; margin-bottom: 10px;}
    .avatar {font-size: 20px;}
""") as demo:
    gr.Markdown("<h1 style='text-align:center;'>💜 Nytheria-3B Q8 Chatbot</h1><p style='text-align:center;'>Sexy UI powered by llama.cpp GGUF</p>")

    chatbot = gr.Chatbot(elem_id="chatbot", bubble_full_width=False, label="Chat with Nytheria")
    msg = gr.Textbox(placeholder="Type your message here...", show_label=False)
    clear = gr.Button("🧹 Clear Chat")

    def respond(message, chat_history):
        chat_history, new_msg = chat(message, chat_history)
        return chat_history, new_msg

    msg.submit(respond, [msg, chatbot], [chatbot, msg])
    clear.click(lambda: [], None, chatbot, queue=False)

demo.launch()