app.py

import gradio as gr
from transformers import AutoTokenizer
from model import MultiHeadClassification
import spaces


model = MultiHeadClassification.from_pretrained(
    'philipp-zettl/multi-head-sequence-classification-model',
    {"GGU": 3, "sentiment": 3}
).to('cuda')
tokenizer = AutoTokenizer.from_pretrained('BAAI/bge-m3')

#@spaces.GPU()
def generate(prompt):
    inputs = tokenizer([prompt]).to('cuda')
    return model(**inputs)


gr.Interface(
    generate,
    inputs="text",
    outputs="label"
)