Closed
Description
Describe the bug
For example when 100 steps are given for num_inference_steps, only 95 are performed.
Reproduction
from diffusers import StableDiffusionXLPipeline
from diffusers.schedulers import DPMSolverMultistepScheduler
import torch
scheduler = DPMSolverMultistepScheduler(
beta_start=0.00085,
beta_end=0.012,
beta_schedule="scaled_linear",
prediction_type="epsilon",
num_train_timesteps=1000,
trained_betas=None,
thresholding=False,
algorithm_type="dpmsolver++",
solver_type="midpoint",
lower_order_final=True,
use_karras_sigmas=True,
)
hf_cache_dir = '/path/to/hf'
pipe = StableDiffusionXLPipeline.from_pretrained(
f"{hf_cache_dir}/stable-diffusion-xl-base-0.9", variant="fp16", use_safetensors=True,
torch_dtype=torch.bfloat16,
)
pipe.enable_model_cpu_offload(0)
pipe.enable_vae_slicing()
pipe.enable_vae_tiling()
pipe.register_modules(scheduler=scheduler)
with torch.no_grad(), torch.amp.autocast('cuda', dtype=torch.bfloat16):
random_generator = torch.Generator()
random_generator.manual_seed(12345)
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt=prompt, output_type="pil",
num_inference_steps=100, generator=random_generator).images[0]
Logs
python test_sdxl2.py
Loading pipeline components...: 100%|███████████████████████████████████████████████████████████████████| 7/7 [00:02<00:00, 2.81it/s]
Loading pipeline components...: 100%|███████████████████████████████████████████████████████████████████| 5/5 [00:03<00:00, 1.65it/s]
95%|███████████████████████████████████████████████████████████████████████████████████████████▏ | 95/100 [00:32<00:01, 2.91it/s]
System Info
Commit b024ebb (main)