import gradio as gr
from transformers import AutoModel, AutoTokenizer
from PIL import Image
import torch

# Load model and tokenizer
tokenizer = AutoTokenizer.from_pretrained('ucaslcl/GOT-OCR2_0', trust_remote_code=True)
model = AutoModel.from_pretrained(
    'ucaslcl/GOT-OCR2_0',
    trust_remote_code=True,
    low_cpu_mem_usage=True,
    device_map='cuda' if torch.cuda.is_available() else 'cpu',
    use_safetensors=True,
    pad_token_id=tokenizer.eos_token_id
)
model = model.eval()
if torch.cuda.is_available():
    model = model.cuda()

# OCR function
def ocr_from_image(image_file, ocr_type):
    if image_file is None:
        return "Please upload an image."

    # Ouvrir le fichier image avec PIL
    image = Image.open(image_file).convert("RGB")
    image_path = "uploaded_image.jpg"
    image.save(image_path)

    # Passer le chemin au modele
    res = model.chat(tokenizer, image_path, ocr_type=ocr_type)
    return res

# OCR types to choose from
ocr_types = ["ocr", "format"]

# Gradio interface
iface = gr.Interface(
    fn=ocr_from_image,
    inputs=[
        gr.File(label="Upload Image", file_types=[".jpg", ".jpeg", ".png"]),
        gr.Radio(ocr_types, label="OCR Type", value="ocr")
    ],
    outputs="text",
    title="?? GOT-OCR2.0 Transformer OCR",
    description="Upload an image file and select the OCR type: plain text (`ocr`) or formatted (`format`)."
)

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