app.py

import torch
import gradio as gr
from transformers import pipeline

# -------------------------------
# Load Model (CPU-safe)
# -------------------------------
pipe = pipeline(
    "text-generation",
    model="TinyLlama/TinyLlama-1.1B-Chat-v1.0",
    torch_dtype=torch.float32,
    device_map=None,
)

# -------------------------------
# Chat Function (messages format)
# -------------------------------
def chat(user_message, history, system_prompt, temperature, max_tokens):
    messages = [{"role": "system", "content": system_prompt}]

    if history:
        messages.extend(history)

    messages.append({"role": "user", "content": user_message})

    prompt = pipe.tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True,
    )

    output = pipe(
        prompt,
        max_new_tokens=int(max_tokens),
        temperature=temperature,
        top_p=0.9,
        do_sample=True,
    )

    assistant_reply = output[0]["generated_text"].split("<|assistant|>")[-1].strip()

    history.append({"role": "user", "content": user_message})
    history.append({"role": "assistant", "content": assistant_reply})

    return history

# -------------------------------
# Gradio UI
# -------------------------------
with gr.Blocks(title="TinyLLaMA Chatbot") as demo:
    gr.Markdown("## 🦙 TinyLLaMA Chatbot")

    preset_prompts = {
        "Pirate 🏴‍☠️": "You are a pirate chatbot. Speak like a pirate.",
        "Teacher 👨‍🏫": "You are a patient teacher.",
        "Coder 👨‍💻": "You are a programming assistant.",
        "Friendly 🤖": "You are a friendly assistant."
    }

    personality = gr.Dropdown(
        choices=list(preset_prompts.keys()),
        value="Pirate 🏴‍☠️",
        label="Choose Personality"
    )

    system_prompt = gr.Textbox(
        value=preset_prompts["Pirate 🏴‍☠️"],
        label="System Prompt"
    )

    personality.change(
        lambda x: preset_prompts[x],
        inputs=personality,
        outputs=system_prompt
    )

    chatbot = gr.Chatbot(type="messages", height=400)
    user_input = gr.Textbox(label="Your Message")

    temperature = gr.Slider(0.1, 1.2, value=0.85)
    max_tokens = gr.Slider(32, 128, value=96, step=16)

    send = gr.Button("Send 🚀")
    clear = gr.Button("Clear 🧹")

    send.click(
        chat,
        inputs=[user_input, chatbot, system_prompt, temperature, max_tokens],
        outputs=chatbot
    )

    user_input.submit(
        chat,
        inputs=[user_input, chatbot, system_prompt, temperature, max_tokens],
        outputs=chatbot
    )

    clear.click(lambda: [], outputs=chatbot)

demo.launch()