app.py

import gradio as gr
from PIL import Image
import torch

# ── Patch transformers BEFORE importing AutoModel ──────────────────────────
from transformers import PretrainedConfig

_original_pretrained_init = PretrainedConfig.__init__

def _patched_pretrained_init(self, *args, **kwargs):
    if not hasattr(self, "forced_bos_token_id"):
        self.forced_bos_token_id = kwargs.get("forced_bos_token_id", None)
    _original_pretrained_init(self, *args, **kwargs)

PretrainedConfig.__init__ = _patched_pretrained_init
# ───────────────────────────────────────────────────────────────────────────

from transformers import AutoProcessor, AutoModelForCausalLM

device = "cuda" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32

print(f"Loading Florence-2 on {device}...")

model = AutoModelForCausalLM.from_pretrained(
    "microsoft/Florence-2-base",
    torch_dtype=torch_dtype,
    trust_remote_code=True,
    low_cpu_mem_usage=False   # ← fix: prevents meta tensor initialization
).to(device)

processor = AutoProcessor.from_pretrained(
    "microsoft/Florence-2-base",
    trust_remote_code=True
)

print("Model loaded successfully!")

TASK_PROMPTS = {
    "Caption": "<CAPTION>",
    "Detailed Caption": "<DETAILED_CAPTION>",
    "More Detailed Caption": "<MORE_DETAILED_CAPTION>",
    "Object Detection": "<OD>",
    "Dense Region Caption": "<DENSE_REGION_CAPTION>",
    "Region Proposal": "<REGION_PROPOSAL>",
    "Caption to Phrase Grounding": "<CAPTION_TO_PHRASE_GROUNDING>",
    "Referring Expression Segmentation": "<REFERRING_EXPRESSION_SEGMENTATION>",
    "OCR": "<OCR>",
    "OCR with Region": "<OCR_WITH_REGION>",
}

TASKS_REQUIRING_TEXT = {"Caption to Phrase Grounding", "Referring Expression Segmentation"}


def run_florence(image: Image.Image, task: str, text_input: str = ""):
    if image is None:
        return "⚠️ Please upload an image."

    task_prompt = TASK_PROMPTS[task]

    if task in TASKS_REQUIRING_TEXT:
        if not text_input.strip():
            return f"⚠️ Task '{task}' requires a text input. Please provide one."
        prompt = task_prompt + text_input.strip()
    else:
        prompt = task_prompt

    try:
        inputs = processor(
            text=prompt,
            images=image,
            return_tensors="pt"
        ).to(device, torch_dtype)

        with torch.no_grad():
            generated_ids = model.generate(
                input_ids=inputs["input_ids"],
                pixel_values=inputs["pixel_values"],
                max_new_tokens=1024,
                num_beams=3,
            )

        generated_text = processor.batch_decode(
            generated_ids, skip_special_tokens=False
        )[0]

        parsed = processor.post_process_generation(
            generated_text,
            task=task_prompt,
            image_size=(image.width, image.height)
        )

        return str(parsed)

    except Exception as e:
        return f"❌ Error during inference: {str(e)}"


def toggle_text_input(task):
    return gr.update(visible=task in TASKS_REQUIRING_TEXT)


with gr.Blocks(title="Florence-2 Demo", theme=gr.themes.Soft()) as demo:
    gr.Markdown(
        """
        # 🖼️ Microsoft Florence-2-base
        Multi-task vision model: captioning, OCR, object detection, segmentation, and more.
        Upload an image, choose a task, and hit **Run**.
        """
    )

    with gr.Row():
        with gr.Column(scale=1):
            image_input = gr.Image(type="pil", label="Input Image")

            task_dropdown = gr.Dropdown(
                choices=list(TASK_PROMPTS.keys()),
                value="Detailed Caption",
                label="Task"
            )

            text_input = gr.Textbox(
                label="Text Input (required for grounding / segmentation tasks)",
                placeholder="e.g. 'a cat on the sofa'",
                visible=False
            )

            run_btn = gr.Button("▶ Run", variant="primary")

        with gr.Column(scale=1):
            output = gr.Textbox(label="Output", lines=20)

    task_dropdown.change(
        fn=toggle_text_input,
        inputs=task_dropdown,
        outputs=text_input
    )

    run_btn.click(
        fn=run_florence,
        inputs=[image_input, task_dropdown, text_input],
        outputs=output
    )

    gr.Examples(
        examples=[
            ["https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/car.jpg", "Detailed Caption", ""],
            ["https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/car.jpg", "More Detailed Caption", ""],
            ["https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/car.jpg", "OCR", ""],
            ["https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/car.jpg", "Object Detection", ""],
            ["https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/car.jpg", "Dense Region Caption", ""],
        ],
        inputs=[image_input, task_dropdown, text_input],
    )

demo.launch()