[fix bug] PixArt inference_steps=1

[lawrence-cj]

This PR fix the bug when PixArt inference with num_inference_steps=1

[hlky]

@bot /style

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[hlky]

@bot /style

[hlky]

Suggested change

	except:
	except Exception:

At minimum due to ruff. What's the exception type raised here? Is there anything else we can check instead of relying on try/except?

[lawrence-cj]

This one is due to the Output difference between

diffusers/src/diffusers/schedulers/scheduling_lcm.py

Lines 48 to 49 in 1001425

	prev_sample: torch.Tensor
	denoised: Optional[torch.Tensor] = None

and

diffusers/src/diffusers/schedulers/scheduling_ddim.py

Lines 46 to 47 in 1001425

	prev_sample: torch.Tensor
	pred_original_sample: Optional[torch.Tensor] = None

What do you think is better to write here?

[lawrence-cj]

gentle ping @hlky

[hlky]

With return_dict=False a tuple is returned, both denoised and pred_original_sample are at index 1 so we can do

latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[1]

[lawrence-cj]

That makes sense. updated.

[hlky]

Thanks @lawrence-cj

[hlky]

Failing tests are unrelated.