app.py

import os
import subprocess
import threading

subprocess.check_call([os.sys.executable, "-m", "pip", "install", "-r", "requirements.txt"])

import spaces
import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer

MODEL_ID = "le-llm/gemma-3-12b-it-reasoning"

SYSTEM_PROMPT = (
    "You are a helpful, concise assistant. Only write replies as the Assistant. Do not invent or continue User messages."
)


def load_model():
    """Lazy-load model & tokenizer (for zeroGPU)."""
    device = "cuda"# if torch.cuda.is_available() else "cpu"
    tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
    model = AutoModelForCausalLM.from_pretrained(
        MODEL_ID,
        torch_dtype=torch.bfloat16 if device == "cuda" else torch.float32,
        device_map="auto" if device == "cuda" else None,
    )
    return model, tokenizer, device


@spaces.GPU
def respond(
    message,
    history: list[dict[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    # Load model/tokenizer each request → allows zeroGPU to cold start & then release
    model, tokenizer, device = load_model()

    # Build conversation
    messages = [{"role": "system", "content": system_message}] + history + [
        {"role": "user", "content": message}
    ]

    input_text = tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True
    )
    inputs = tokenizer(input_text, return_tensors="pt").to(device)

    # Streamer setup
    streamer = TextIteratorStreamer(
        tokenizer,
        skip_special_tokens=True,
        skip_prompt=True
    )

    # Run model.generate in background thread
    generation_kwargs = dict(
        **inputs,
        max_new_tokens=max_tokens,
        temperature=temperature,
        top_p=top_p,
        do_sample=True,
        eos_token_id=tokenizer.eos_token_id,
        streamer=streamer,
    )
    thread = threading.Thread(target=model.generate, kwargs=generation_kwargs)
    thread.start()

    # Yield tokens as they come in
    partial_output = ""
    for new_text in streamer:
        partial_output += new_text
        yield partial_output


chatbot = gr.ChatInterface(
    respond,
    type="messages",
    additional_inputs=[
        gr.Textbox(value=SYSTEM_PROMPT, label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95,
            step=0.05,
            label="Top-p (nucleus sampling)",
        ),
    ],
)

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