app.py

from transformers import AutoProcessor, AutoModelForImageTextToText, MllamaForConditionalGeneration
import torch
from threading import Thread
import gradio as gr
import spaces
from PIL import Image


# from zipnn import zipnn_hf

# zipnn_hf()

# model_id = "royleibov/Llama-3.2-11B-Vision-Instruct-ZipNN-Compressed"
model_id = "unsloth/Llama-3.2-11B-Vision"

model = AutoModelForImageTextToText.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
processor = AutoProcessor.from_pretrained(model_id)


@spaces.GPU
def score_it(input_img):
    # Convert numpy array to PIL Image
    pil_image = Image.fromarray(input_img.astype('uint8'))
    
    # Now use the PIL Image as before
    image = pil_image.convert("RGB").resize((224, 224))

    prompt = "<|begin_of_text|><|image|this script says"
    inputs = processor(image, prompt, return_tensors="pt").to(model.device)

    output = model.generate(**inputs, max_new_tokens=200)
    return processor.decode(output[0])
    


demo = gr.Interface(fn=score_it, title="Upload your English script and get the score",
                        inputs=[gr.Image()],
                        outputs=['text'],
                        stop_btn="Stop Generation",
                        )

demo.launch(debug=True)