app.py

import gradio as gr
import keras_nlp
print("Modules loaded!")

print("Fetching model...")
model = keras_nlp.models.GemmaCausalLM.from_preset("hf://bhashwarsengupta/gemma2-instruct-2b-en-finance")
print("model successfully loaded!")

def respond(
    message,
    history: list[tuple[str, str]]
):

    messages = f"Question:\n{message}\n\nAnswer:\n"

    print("Generating response...")
    output = model.generate(messages)
    print("Response generated!")

    # Split by "Answer:" from the right and get the last part
    response = output.rsplit("Answer:\n", 1)[-1]
    
    return response 


demo = gr.ChatInterface(
    respond
)

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