[Community] CPU Under Utilized #834
Description
Describe the bug
When using CPU as device, the CPU is very under utilized. I get about 50-66% utilization. The fans don't even need to turn on. Should I be running a special version of PyTorch, or?
Reproduction
# make sure you're logged in with `huggingface-cli login`
from diffusers import StableDiffusionPipeline
import subprocess, random, time, torch
prompt = "A beautiful 3d matte portrait of a _identity-adult_ warrior by Ilya Kuvshinov in a _color_ color scheme, featured on ArtStation"
neg_prompt = ""
width = 512
height = 640
steps = 50
cfg = 13.5
batches = 5
seed = -1
device = 'cpu'
print('Setting up Diffusion Pipeline...')
pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", revision="fp16", use_auth_token=True)
pipe = pipe.to(device)
pipe.enable_attention_slicing() # Tried on and off, seems to be no real difference in time of about 22m a image on Ryzen 5 2600
print('Pipeline setup complete.')
if seed == -1:
seed = random.randint(0, 9999999999)
for batch in range(batches):
if batch != 0:
seed += 1
gen_seed = torch.Generator(device).manual_seed(seed)
print(f'Starting diffusion run {batch} of {batches} ...\n')
print('PROMPT:')
print(prompt+"\n")
if neg_prompt != '':
print('NEGATIVE PROMPT:')
print(neg_prompt+"\n")
image = pipe(prompt=prompt, negative_prompt=neg_prompt, num_inference_steps=steps, width=width, height=height, guidance_scale=cfg, generator=gen_seed).images[0]
print('Diffusion Complete!')
file = str(time.time())+"_result.png"
print(f'Saving result as \'{file}\'\n')
image.save(file)Logs
No response
System Info
diffusersversion: 0.4.2- Platform: Windows-10-10.0.22621-SP0
- Python version: 3.10.2
- PyTorch version (GPU?): 1.11.0+cpu (False)
- Huggingface_hub version: 0.10.1
- Transformers version: 4.23.1
- Using GPU in script?: False
- Using distributed or parallel set-up in script?: False