app.py

import gradio as gr
from transformers import AutoProcessor, AutoModelForCausalLM
import spaces

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

models = {
    'J-LAB/Florence_2_B_FluxiAI_Product_Caption': AutoModelForCausalLM.from_pretrained('J-LAB/Florence_2_B_FluxiAI_Product_Caption', trust_remote_code=True).to("cuda").eval(),
    'J-LAB/Florence_2_L_FluxiAI_Product_Caption': AutoModelForCausalLM.from_pretrained('J-LAB/Florence_2_L_FluxiAI_Product_Caption', trust_remote_code=True).to("cuda").eval()
}

processors = {
    'J-LAB/Florence_2_B_FluxiAI_Product_Caption': AutoProcessor.from_pretrained('J-LAB/Florence_2_B_FluxiAI_Product_Caption', trust_remote_code=True),
    'J-LAB/Florence_2_L_FluxiAI_Product_Caption': AutoProcessor.from_pretrained('J-LAB/Florence_2_L_FluxiAI_Product_Caption', trust_remote_code=True)
}

DESCRIPTION = "# [Florence-2 Product Describe by Fluxi IA](https://huggingface.co/microsoft/Florence-2-large)"

@spaces.GPU
def run_example(task_prompt, image, text_input=None, model_id='J-LAB/Florence-Idesire'):
    model = models[model_id]
    processor = processors[model_id]
    if text_input is None:
        prompt = task_prompt
    else:
        prompt = task_prompt + text_input
    inputs = processor(text=prompt, images=image, return_tensors="pt").to("cuda")
    generated_ids = 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 = processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
    parsed_answer = processor.post_process_generation(
        generated_text,
        task=task_prompt,
        image_size=(image.width, image.height)
    )
    return parsed_answer
    
def process_image(image, task_prompt, text_input=None, model_id='J-LAB/Florence_2_B_FluxiAI_Product_Caption'):
    image = Image.fromarray(image)  # Convert NumPy array to PIL Image
    if task_prompt == 'Product Caption':
        task_prompt = '<PC>'
        results = run_example(task_prompt, image, model_id=model_id)
    elif task_prompt == 'More Detailed Caption':
        task_prompt = '<MORE_DETAILED_CAPTION>'
        results = run_example(task_prompt, image, model_id=model_id)
    else:
        return "", None  # Return empty string and None for unknown task prompts

    # Remove the key and get the text value
    if results and task_prompt in results:
        output_text = results[task_prompt]
    else:
        output_text = ""

    # Convert newline characters to HTML line breaks
    output_text = output_text.replace("\n\n", "<br><br>").replace("\n", "<br>")

    return output_text

css = """
  #output {
    height: 500px; 
    overflow: auto; 
    border: 1px solid #ccc; 
  }
"""

single_task_list = [
    'Product Caption', 'More Detailed Caption'
]

with gr.Blocks(css=css) as demo:
    gr.Markdown(DESCRIPTION)
    with gr.Tab(label="Florence-2 Image Captioning"):
        with gr.Row():
            with gr.Column():
                input_img = gr.Image(label="Input Picture")
                model_selector = gr.Dropdown(choices=list(models.keys()), label="Model", value='J-LAB/Florence_2_B_FluxiAI_Product_Caption')
                task_type = gr.Radio(choices=['Single task', 'Cascased task'], label='Task type selector', value='Single task')
                task_prompt = gr.Dropdown(choices=single_task_list, label="Task Prompt", value="Caption")
                text_input = gr.Textbox(label="Text Input (optional)")
                submit_btn = gr.Button(value="Submit")
            with gr.Column():
                output_text = gr.HTML(label="Output Text")

        submit_btn.click(process_image, [input_img, task_prompt, text_input, model_selector], [output_text])

demo.launch(debug=True)