app.py

import gradio as gr
import torch
import os
import tempfile
from PIL import Image, ImageOps

from transformers import AutoProcessor, AutoModelForImageTextToText

MODEL_PATH = "zai-org/GLM-OCR"

processor = AutoProcessor.from_pretrained(MODEL_PATH, trust_remote_code=True)
model = AutoModelForImageTextToText.from_pretrained(
    pretrained_model_name_or_path=MODEL_PATH,
    torch_dtype=torch.float32,
    device_map="cpu",
    trust_remote_code=True,
)
model.eval()

TASK_PROMPTS = {
    "Text": "Text Recognition:",
    "Formula": "Formula Recognition:",
    "Table": "Table Recognition:",
}


def process_image(image, task):
    if image is None:
        return "Please upload an image first.", "Please upload an image first."

    if image.mode in ("RGBA", "LA", "P"):
        image = image.convert("RGB")
    image = ImageOps.exif_transpose(image)

    tmp = tempfile.NamedTemporaryFile(delete=False, suffix=".png")
    image.save(tmp.name, "PNG")
    tmp.close()

    prompt = TASK_PROMPTS.get(task, "Text Recognition:")

    messages = [
        {
            "role": "user",
            "content": [
                {"type": "image", "url": tmp.name},
                {"type": "text", "text": prompt},
            ],
        }
    ]

    inputs = processor.apply_chat_template(
        messages,
        tokenize=True,
        add_generation_prompt=True,
        return_dict=True,
        return_tensors="pt",
    ).to("cpu")

    inputs.pop("token_type_ids", None)

    with torch.no_grad():
        generated_ids = model.generate(**inputs, max_new_tokens=4096)

    output_text = processor.decode(
        generated_ids[0][inputs["input_ids"].shape[1]:],
        skip_special_tokens=True,
    )

    os.unlink(tmp.name)

    result = output_text.strip()
    return result, result


with gr.Blocks(title="GLM-OCR CPU") as demo:
    gr.Markdown("# GLM-OCR (CPU)")
    with gr.Row():
        with gr.Column(scale=1):
            image_input = gr.Image(
                type="pil",
                label="Upload Image",
                sources=["upload", "clipboard"],
                height=300,
            )
            task = gr.Radio(
                choices=list(TASK_PROMPTS.keys()),
                value="Text",
                label="Recognition Type",
            )
            btn = gr.Button("Perform OCR", variant="primary")
        with gr.Column(scale=2):
            output_text = gr.Textbox(label="Raw Output", interactive=True, lines=18)
            with gr.Accordion("Rendered Markdown", open=False):
                output_md = gr.Markdown(label="Rendered Markdown")

    btn.click(
        fn=process_image,
        inputs=[image_input, task],
        outputs=[output_text, output_md],
    )

    image_input.change(
        fn=lambda: ("", ""),
        inputs=None,
        outputs=[output_text, output_md],
    )

if __name__ == "__main__":
    demo.queue(max_size=50).launch(
        server_name="0.0.0.0", server_port=7860, show_error=True,
        theme="NoCrypt/miku")