app.py

import gradio as gr
import subprocess
import torch
from PIL import Image
from transformers import AutoProcessor, AutoConfig
import importlib.util, sys, os

subprocess.run(
    "pip install --upgrade transformers>=4.50.0",
    shell=True,
    check=True
)

model_id = "microsoft/Florence-2-base-ft"

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

config = AutoConfig.from_pretrained(model_id, trust_remote_code=True)

config_mod_name = config.__class__.__module__ 
config_mod      = sys.modules[config_mod_name]
code_dir        = os.path.dirname(config_mod.__file__)

modeling_path = os.path.join(code_dir, "modeling_florence2.py")
if not os.path.exists(modeling_path):
    raise FileNotFoundError(f"Couldn’t find {modeling_path}")

spec = importlib.util.spec_from_file_location("florence2_modeling", modeling_path)
flor_mod = importlib.util.module_from_spec(spec)
sys.modules["florence2_modeling"] = flor_mod
spec.loader.exec_module(flor_mod)

FlorenceLM = flor_mod.Florence2LanguageForConditionalGeneration
florence_model = FlorenceLM.from_pretrained(
    model_id,
    trust_remote_code=True
).to(device).eval()
florence_processor = AutoProcessor.from_pretrained(model, trust_remote_code=True)

def generate_caption(image):
    if not isinstance(image, Image.Image):
        image = Image.fromarray(image)
    
    inputs = florence_processor(text="<MORE_DETAILED_CAPTION>", images=image, return_tensors="pt")
    inputs = {k: v.to(device) for k, v in inputs.items()}
    generated_ids = florence_model.generate(
        input_ids=inputs["input_ids"],
        pixel_values=inputs["pixel_values"],
        max_new_tokens=1024,
        early_stopping=False,
        do_sample=False,
        num_beams=3,
    )
    generated_text = florence_processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
    parsed_answer = florence_processor.post_process_generation(
        generated_text,
        task="<MORE_DETAILED_CAPTION>",
        image_size=(image.width, image.height)
    )
    prompt =  parsed_answer["<MORE_DETAILED_CAPTION>"]
    print("\n\nGeneration completed!:"+ prompt)
    return prompt
    
demo = gr.Interface(generate_caption,
                  inputs=[gr.Image(label="Input Image")],
                  outputs = [gr.Textbox(label="Output Prompt", lines=3, show_copy_button = True),
                            ],
                  theme="Yntec/HaleyCH_Theme_Orange",
                 )
demo.launch(debug=True)