Apply suggestions from code review

Author: patrickvonplaten

docs/source/en/optimization/fp16.mdx

@@ -162,9 +162,7 @@ images = pipe([prompt], width=3840, height=2224).images
 The output image will have some tile-to-tile tone variation from the tiles having separate decoders, but you shouldn't see sharp seams between the tiles. The tiling is turned off for images that are 512x512 or smaller.
 
 
-=======
 <a name="sequential_offloading"></a>
->>>>>>> 664b4de9e22a825875b6bad45867f8f77cdf95d6:docs/source/en/optimization/fp16.mdx
 ## Offloading to CPU with accelerate for memory savings
 
 For additional memory savings, you can offload the weights to CPU and only load them to GPU when performing the forward pass.

tests/pipelines/stable_diffusion/test_stable_diffusion.py

@@ -422,7 +422,7 @@ def test_stable_diffusion_vae_slicing(self):
     def test_stable_diffusion_vae_tiling(self):
         device = "cpu"  # ensure determinism for the device-dependent torch.Generator
         unet = self.dummy_cond_unet
-        scheduler = LMSDiscreteScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear")
+        scheduler = LMSDiscreteScheduler.from_pretrained("CompVis/stable-diffusion-v1-4")
         vae = self.dummy_vae
         bert = self.dummy_text_encoder
         tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip")