app.py

import gradio as gr
from huggingface_hub import InferenceClient


def respond(
    message,
    history: list[dict[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
    hf_token: gr.OAuthToken,
):
    """
    Uses the Hugging Face InferenceClient with a token (OAuth) to access the model.
    This works with any text-to-text model like BART, T5, Pegasus, etc.
    """
    client = InferenceClient(
        token=hf_token.token,
        model="Bocklitz-Lab/lit2vec-tldr-bart-model"
    )

    # Construct input text (optionally prepend system message)
    full_input = f"{system_message.strip()}\n\n{message.strip()}"

    # Generate full response in one call
    response = client.text_generation(
        full_input,
        max_new_tokens=max_tokens,
        temperature=temperature,
        top_p=top_p,
        stream=False  # Set to True for token streaming
    )

    yield response


chatbot = gr.ChatInterface(
    respond,
    chatbot=gr.Chatbot(),
    textbox=gr.Textbox(placeholder="Enter text to summarize...", container=False, scale=7),
    additional_inputs=[
        gr.Textbox(value="Summarize the following scientific text.", label="System message"),
        gr.Slider(minimum=16, maximum=1024, value=256, step=8, 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"),
    ],
)

with gr.Blocks() as demo:
    with gr.Sidebar():
        gr.LoginButton()
    chatbot.render()


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