app.py

import gradio as gr
import spaces

# Initialize the model only once, outside of any function
# Ensure that CUDA initialization happens within the worker process
model_pipe = None

@spaces.GPU
def generate(model_name, image, text):
    global model_pipe
    import torch
    torch.jit.script = lambda f: f
    
    from t2v_metrics import VQAScore, list_all_vqascore_models
    
    if model_pipe is None:
        print("Initializing model...")
        model_pipe = VQAScore(model="clip-flant5-xl", device="cuda")  # our recommended scoring model
        # model_pipe.to("cuda")
    
    print(list_all_vqascore_models())
    print("Image:", image)
    print("Text:", text)
    
    print("Generating!")
    result = model_pipe(images=[image], texts=[text])
    return result

iface = gr.Interface(
    fn=generate,  # function to call
    inputs=[gr.Dropdown(["clip-flant5-xl", "clip-flant5-xxl"], label="Model Name"), gr.Image(type="filepath"), gr.Textbox(label="Prompt")],  # define the types of inputs
    outputs="number",  # define the type of output
    title="VQAScore",  # title of the app
    description="This model evaluates the similarity between an image and a text prompt."
).launch()