app.py

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

model_id = "Qwen/Qwen3-1.7B"
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype = "auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_id)

if torch.cuda.is_available():
    device = torch.device("cuda")
else:
    device = torch.device("cpu")

model = model.to(device)

@spaces.GPU
def respuesta(
    message,
    history,
    system_message,
    max_tokens,
    temperature,
    top_p
):
    messages = [{"role": "system", "content": system_message}]

    for val in history:
        if val[0]:
            messages.append({"role": "user", "content": val[0]})
        if val[1]:
            messages.append({"role": "assistant", "content": val[1]})

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

    input_ids = tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True,
        enable_thinking=True
    ).to(model.device)

    model_inputs = tokenizer([text], return_tensor='pt')

    outputs = model.generate(
        **model_inputs,
        max_new_tokens=max_tokens,
        do_sample=True,
        temperature=temperature,
        top_p=top_p
    )

    response = ''

    for message in tokenizer.decode(
        outputs[0][input_ids.shape[-1]:],
        skip_special_tokens=True
    ):
        response += message
        yield response


demo = gr.ChatInterface(
    respuesta,
    additional_inputs=[
        gr.Textbox(value="Eres un chatbot amigable", label="System messaage"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4, 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"),            
    ]
)

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