app.py

import gradio as gr
import numpy as np
import random
import torch
from diffusers import DiffusionPipeline
from accelerate import init_empty_weights, load_checkpoint_and_dispatch

# Configuración para usar bfloat16 y CUDA si está disponible
dtype = torch.bfloat16
device = "cuda" if torch.cuda.is_available() else "cpu"

# Inicializar el modelo solo una vez y cargarlo en RAM y GPU
pipe = None

def load_model():
    global pipe
    if pipe is None:
        with init_empty_weights():
            pipe = DiffusionPipeline.from_pretrained(
                "black-forest-labs/FLUX.1-schnell", 
                torch_dtype=dtype
            )
        # Despachar los pesos a la GPU (sin "named_parameters")
        pipe = load_checkpoint_and_dispatch(
            pipe, 
            "black-forest-labs/FLUX.1-schnell", 
            device_map="auto",  # Automatiza el uso de RAM y GPU
            offload_folder=None,  # Evita que se almacenen los pesos temporalmente en el disco
        ).to(device)

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 2048

@spaces.GPU()
def infer(prompt, seed=42, randomize_seed=False, width=1024, height=1024, num_inference_steps=4, num_images=1, progress=gr.Progress(track_tqdm=True)):
    load_model()  # Asegurarse de que el modelo esté cargado antes de la inferencia
    
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    generator = torch.Generator().manual_seed(seed)
    
    images = []
    for _ in range(num_images):
        image = pipe(
                prompt=prompt, 
                width=width,
                height=height,
                num_inference_steps=num_inference_steps, 
                generator=generator,
                guidance_scale=0.0
        ).images[0]
        images.append(image)
    
    return images, seed

examples = [
    "a tiny astronaut hatching from an egg on the moon",
    "a cat holding a sign that says hello world",
    "an anime illustration of a wiener schnitzel",
]

css = """
#col-container {
    margin: 0 auto;
    max-width: 520px;
}
"""

with gr.Blocks(css=css) as demo:
    
    with gr.Column(elem_id="col-container"):
        gr.Markdown(f"""# FLUX.1 [schnell]
12B param rectified flow transformer distilled from [FLUX.1 [pro]](https://blackforestlabs.ai/) for 4 step generation
[[blog](https://blackforestlabs.ai/announcing-black-forest-labs/)] [[model](https://huggingface.co/black-forest-labs/FLUX.1-schnell)]
        """)
        
        with gr.Row():
            
            prompt = gr.Text(
                label="Prompt",
                show_label=False,
                max_lines=1,
                placeholder="Enter your prompt",
                container=False,
            )
            
            run_button = gr.Button("Run", scale=0)
        
        results = gr.Gallery(label="Results", show_label=False, elem_id="image-gallery")
        
        with gr.Accordion("Advanced Settings", open=False):
            
            seed = gr.Slider(
                label="Seed",
                minimum=0,
                maximum=MAX_SEED,
                step=1,
                value=0,
            )
            
            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
            
            with gr.Row():
                
                width = gr.Slider(
                    label="Width",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1024,
                )
                
                height = gr.Slider(
                    label="Height",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1024,
                )
            
            with gr.Row():
                
                num_inference_steps = gr.Slider(
                    label="Number of inference steps",
                    minimum=1,
                    maximum=50,
                    step=1,
                    value=4,
                )
            
            num_images = gr.Slider(
                label="Number of images",
                minimum=1,
                maximum=300,
                step=1,
                value=1,
            )
        
        gr.Examples(
            examples = examples,
            fn = infer,
            inputs = [prompt],
            outputs = [results, seed],
            cache_examples="lazy"
        )

    gr.on(
        triggers=[run_button.click, prompt.submit],
        fn = infer,
        inputs = [prompt, seed, randomize_seed, width, height, num_inference_steps, num_images],
        outputs = [results, seed]
    )

# Crear un enlace público con share=True
demo.launch(share=True)