app.py

import gradio as gr
from random import randint
from all_models import models
from externalmod import gr_Interface_load, randomize_seed
import asyncio
import os
from threading import RLock

# Create a lock for thread safety
lock = RLock()

# Load Hugging Face token from environment variables
HF_TOKEN = os.environ.get("HF_TOKEN")

# Function to load models
def load_fn(models):
    global models_load
    models_load = {}
    for model in models:
        if model not in models_load:
            try:
                print(f"Loading model: {model}")
                m = gr_Interface_load(f'models/{model}', hf_token=HF_TOKEN)
                print(f"Successfully loaded: {model}")
            except Exception as error:
                print(f"Error loading {model}: {error}")
                m = gr.Interface(lambda: None, ['text'], ['image'])
            models_load[model] = m

# Load models
print("Loading models...")
load_fn(models)
print("Models loaded successfully.")

num_models = min(3, len(models))  # Reduce to 3 models to prevent GPU overloading
starting_seed = randint(1941, 2024)
print(f"Starting seed: {starting_seed}")

# Extend choices to match num_models
def extend_choices(choices):
    extended = choices[:num_models] + (num_models - len(choices[:num_models])) * ['NA']
    return extended

# Update image boxes based on selected models
def update_imgbox(choices):
    choices_extended = extend_choices(choices)
    return [gr.Image(None, label=m, visible=(m != 'NA')) for m in choices_extended]

# Asynchronous inference function
async def infer(model_str, prompt, seed=1, timeout=600):
    if model_str == 'NA':
        return None
    try:
        print(f"Running inference on {model_str} with prompt: '{prompt}'")
        task = asyncio.to_thread(models_load[model_str].fn, prompt=prompt, seed=seed, token=HF_TOKEN)
        result = await asyncio.wait_for(task, timeout=timeout)
        if result:
            with lock:
                image_path = "image.png"
                result.save(image_path)
                return image_path
    except Exception as e:
        print(f"Error in inference for {model_str}: {e}")
    return None

# Wrapper function for inference
def gen_fnseed(model_str, prompt, seed=1):
    if model_str == 'NA':
        return None
    return asyncio.run(infer(model_str, prompt, seed))

# Create Gradio interface
print("Creating Gradio interface...")
with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
    gr.HTML("<center><h1>Compare-3</h1></center>")
    with gr.Tab('Compare-3'):
        txt_input = gr.Textbox(label='Your prompt:', lines=4)
        gen_button = gr.Button('Generate images')
        seed = gr.Slider(label="Seed (max 3999999999)", minimum=0, maximum=3999999999, step=1, value=starting_seed)
        seed_rand = gr.Button("Randomize Seed 🎲")
        seed_rand.click(randomize_seed, None, [seed])
        
        output = [gr.Image(label=m) for m in models[:num_models]]
        current_models = [gr.Textbox(m, visible=False) for m in models[:num_models]]
        
        for m, o in zip(current_models, output):
            gen_button.click(gen_fnseed, inputs=[m, txt_input, seed], outputs=[o])
        
        with gr.Accordion('Model selection'):
            model_choice = gr.CheckboxGroup(models, label=f'Choose up to {num_models} models', value=models[:num_models])
            model_choice.change(update_imgbox, model_choice, output)
            model_choice.change(extend_choices, model_choice, current_models)

# Reduce concurrency to avoid T4 overload
demo.queue(default_concurrency_limit=50, max_size=100)
print("Launching Gradio interface...")
demo.launch(show_api=False, max_threads=50, debug=True)