app.py

import gradio as gr

from transformers import AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained("automaise/quokka-7b")

def create_gradio_app(model, tokenizer):
    def gradio_fn(question):
        inputs = tokenizer(question, return_tensors="pt")
        outputs = model(inputs)[0]
        generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
        return generated_text

    iface = gr.Interface(
        fn=gradio_fn,
        inputs=gr.inputs.Textbox(),
        outputs=gr.outputs.Textbox(),
        live=True,
        title="Gradio App",
        description="Create a gradio app using the Quokka-7b model.",
    )
    return iface

iface = create_gradio_app(model, tokenizer)

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