#!/usr/bin/env python3
from diffusers import StableDiffusionPipeline, UniPCMultistepScheduler
import time
import torch
import sys

path = sys.argv[1]
use_device_map = bool(int(sys.argv[2]))

start_time = time.time()

if use_device_map:
    print("Load directly on GPU")
    pipe = StableDiffusionPipeline.from_pretrained(path, torch_dtype=torch.float16, device_map="auto")
else:
    print("Load directly on CPU")
    pipe = StableDiffusionPipeline.from_pretrained(path, torch_dtype=torch.float16)
    pipe = pipe.to("cuda")

pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)

prompt = "a highly realistic photo of green turtle"

print("Loading Time", time.time() - start_time)
generator = torch.Generator(device="cuda").manual_seed(0)
image = pipe(prompt, generator=generator, num_inference_steps=15).images[0]
print("Time", time.time() - start_time)