Support LMSDiscreteScheduler in LDMPipeline (#891)

* Support LMSDiscreteScheduler in LDMPipeline

This is a small change to support all schedulers such as LMSDiscreteScheduler in LDMPipeline.

What's changed
-------
* Add the `scale_model_input` function before `step` to ensure correct denoising (L77)

* Add "scale the initial noise by the standard deviation required by the scheduler"

* run `make style`
Co-authored-by: Anton Lozhkov <anton@huggingface.co>

src/diffusers/pipelines/latent_diffusion_uncond/pipeline_latent_diffusion_uncond.py

@@ -64,6 +64,9 @@ class LDMPipeline(DiffusionPipeline):
         )
         latents = latents.to(self.device)
 
+        # scale the initial noise by the standard deviation required by the scheduler
+        latents = latents * self.scheduler.init_noise_sigma
+
         self.scheduler.set_timesteps(num_inference_steps)
 
         # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
@@ -74,8 +77,9 @@ class LDMPipeline(DiffusionPipeline):
             extra_kwargs["eta"] = eta
 
         for t in self.progress_bar(self.scheduler.timesteps):
+            latent_model_input = self.scheduler.scale_model_input(latents, t)
             # predict the noise residual
-            noise_prediction = self.unet(latents, t).sample
+            noise_prediction = self.unet(latent_model_input, t).sample
             # compute the previous noisy sample x_t -> x_t-1
             latents = self.scheduler.step(noise_prediction, t, latents, **extra_kwargs).prev_sample