[Bug]: Training with Rescale Noise on and sampling with anything but DDIM fails.

[O-J1]

What happened?

Issue ported from the now closed bug report forums on Discord.

1. Turn on rescale noise
2. Set sampling with any sampler not DDIDM
3. Start training and observe it failing.

This is a known (and old) but niche bug.

What did you expect would happen?

For it to not fail when using anything but DDIM

Relevant log output

: 100%|███████████████████████████████████████████| 16/16 [00:46<00:00,  2.93s/it, loss=0.0571, smooth loss=0.0512]
step: 100%|███████████████████████████████████████████| 16/16 [00:44<00:00,  2.76s/it, loss=0.0403, smooth loss=0.0512]
step: 100%|████████████████████████████████████████████| 16/16 [00:43<00:00,  2.70s/it, loss=0.0693, smooth loss=0.052]
step: 100%|████████████████████████████████████████████| 16/16 [00:43<00:00,  2.70s/it, loss=0.0555, smooth loss=0.052]
step: 100%|███████████████████████████████████████████| 16/16 [00:43<00:00,  2.70s/it, loss=0.0577, smooth loss=0.0522]
step: 100%|███████████████████████████████████████████| 16/16 [00:43<00:00,  2.69s/it, loss=0.0605, smooth loss=0.0519]
step: 100%|███████████████████████████████████████████| 16/16 [00:43<00:00,  2.70s/it, loss=0.0496, smooth loss=0.0518]
step: 100%|███████████████████████████████████████████| 16/16 [00:43<00:00,  2.74s/it, loss=0.0612, smooth loss=0.0519]
step: 100%|███████████████████████████████████████████| 16/16 [00:44<00:00,  2.79s/it, loss=0.0532, smooth loss=0.0524]
step: 100%|█████████████████████████████████████████████| 16/16 [00:44<00:00,  2.76s/it, loss=0.05, smooth loss=0.0521]
sampling:  97%|█████████████████████████████████████████████████████████████████████▋  | 30/31 [00:07<00:00,  3.89it/s]
Traceback (most recent call last):                                                              | 0/16 [00:00<?, ?it/s]
  File "C:\repos\OneTrainer\modules\trainer\GenericTrainer.py", line 254, in __sample_loop0/31 [00:07<00:00,  4.08it/s]
    self.model_sampler.sample(
  File "C:\repos\OneTrainer\modules\modelSampler\StableDiffusionXLSampler.py", line 471, in sample
    image = self.__sample_base(
  File "C:\repos\OneTrainer\venv\lib\site-packages\torch\utils\_contextlib.py", line 116, in decorate_context
    return func(*args, **kwargs)
  File "C:\repos\OneTrainer\modules\modelSampler\StableDiffusionXLSampler.py", line 171, in __sample_base
    latent_image = noise_scheduler.step(
  File "C:\repos\OneTrainer\venv\src\diffusers\src\diffusers\schedulers\scheduling_euler_ancestral_discrete.py", line 415, in step
    sigma_to = self.sigmas[self.step_index + 1]
IndexError: index 31 is out of bounds for dimension 0 with size 31
Error during sampling, proceeding without sampling
sampling: 100%|████████████████████████████████████████████████████████████████████████| 31/31 [00:07<00:00,  4.01it/s]
step: 100%|███████████████████████████████████████████| 16/16 [01:06<00:00,  4.18s/it, loss=0.0376, smooth loss=0.0525]

Output of pip freeze

N/A Occurs in every single version of OT.