app.py

import gradio as gr
import subprocess
import torch
from PIL import Image
from transformers import AutoProcessor, AutoModelForConditionalGeneration

subprocess.run(
    'pip install flash-attn --no-build-isolation',
    env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"},
    shell=True
)

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = 'microsoft/Florence-2-base-ft'
florence_model = AutoModelForConditionalGeneration.from_pretrained(
    model,
    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)