app.py

import gradio as gr
import numpy as np
import random
from diffusers import DiffusionPipeline
import torch
import time
import psutil

# Get the number of physical CPU cores (excluding hyperthreads)
NUM_CPU_CORES = psutil.cpu_count(logical=True)

# Cap the number of threads to the available physical cores
MAX_THREADS = min(8, NUM_CPU_CORES)

# Device and hardware configuration
DEVICE = "cpu"

# Model Options (optimized for CPU and memory constraints)
MODEL_OPTIONS = {
    "Medium Quality (Faster)": "stabilityai/stable-diffusion-2-base",
    "Fastest (Draft Quality)": "hf-internal-testing/tiny-stable-diffusion-pipe",
}

# Default to fastest model and lower image size for limited resources
DEFAULT_MODEL_ID = MODEL_OPTIONS["Fastest (Draft Quality)"]
DEFAULT_IMAGE_SIZE = 512  # Lower default resolution

# Cache models to avoid reloading
PIPELINES = {}

def load_pipeline(model_id):
    if model_id in PIPELINES:
        return PIPELINES[model_id]
    else:
        pipe = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float32)
        pipe.to(DEVICE)
        PIPELINES[model_id] = pipe
        return pipe

def generate_image(prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, num_images, model_choice):
    if not prompt:
        raise gr.Error("Будь ласка, введіть опис для зображення.")

    torch.set_num_threads(MAX_THREADS)  # Set the maximum number of threads

    pipe = load_pipeline(MODEL_OPTIONS[model_choice])

    # Adjust memory usage based on available RAM
    torch.cuda.empty_cache()  

    generator = torch.Generator(device=DEVICE)
    if not randomize_seed:
        generator = generator.manual_seed(seed)

    start_time = time.time()
    images = pipe(
        prompt,
        negative_prompt=negative_prompt,
        width=width,
        height=height,
        guidance_scale=guidance_scale,
        num_inference_steps=num_inference_steps,
        num_images_per_prompt=num_images,
        generator=generator,
    ).images

    end_time = time.time()
    generation_time = end_time - start_time

    return images, f"Час генерації: {generation_time:.2f} секунд"

# ... (Gradio interface remains the same)

    generation_time = end_time - start_time

    return images, f"Час генерації: {generation_time:.2f} секунд"

    run_button = gr.Button("Згенерувати")
    gallery = gr.Gallery(label="Згенеровані зображення")
    status_text = gr.Textbox(label="Статус") 

    run_button.click(
        fn=generate_image,
        inputs=[prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, num_images, model_choice],
        outputs=[gallery, status_text],  # Output both the gallery and status text
    )