[Docs] Advertise fp16 instead of autocast (#740)

up

README.md

@@ -74,11 +74,14 @@ You need to accept the model license before downloading or using the Stable Diff
 
 ### Text-to-Image generation with Stable Diffusion
 
+We recommend using the model in [half-precision (`fp16`)](https://pytorch.org/blog/accelerating-training-on-nvidia-gpus-with-pytorch-automatic-mixed-precision/) as it gives almost always the same results as full
+precision while being roughly twice as fast and requiring half the amount of GPU RAM.
+
 ```python
 # make sure you're logged in with `huggingface-cli login`
 from diffusers import StableDiffusionPipeline
 
-pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4")
+pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", torch_type=torch.float16, revision="fp16")
 pipe = pipe.to("cuda")
 
 prompt = "a photo of an astronaut riding a horse on mars"
@@ -105,8 +108,8 @@ prompt = "a photo of an astronaut riding a horse on mars"
 image = pipe(prompt).images[0]  
 ```
 
-If you are limited by GPU memory, you might want to consider using the model in `fp16` as 
-well as chunking the attention computation.
+If you are limited by GPU memory, you might want to consider chunking the attention computation in addition 
+to using `fp16`.
 The following snippet should result in less than 4GB VRAM.
 
 ```python
@@ -122,7 +125,7 @@ pipe.enable_attention_slicing()
 image = pipe(prompt).images[0]  
 ```
 
-Finally, if you wish to use a different scheduler, you can simply instantiate
+If you wish to use a different scheduler, you can simply instantiate
 it before the pipeline and pass it to `from_pretrained`.
     
 ```python
@@ -148,6 +151,24 @@ image = pipe(prompt).images[0]
 image.save("astronaut_rides_horse.png")
 ```
 
+If you want to run Stable Diffusion on CPU or you want to have maximum precision on GPU, 
+please run the model in the default *full-precision* setting:
+
+```python
+# make sure you're logged in with `huggingface-cli login`
+from diffusers import StableDiffusionPipeline
+
+pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4")
+
+# disable the following line if you run on CPU
+pipe = pipe.to("cuda")
+
+prompt = "a photo of an astronaut riding a horse on mars"
+image = pipe(prompt).images[0]  
+    
+image.save("astronaut_rides_horse.png")
+```
+
 ### Image-to-Image text-guided generation with Stable Diffusion
 
 The `StableDiffusionImg2ImgPipeline` lets you pass a text prompt and an initial image to condition the generation of new images.

docs/source/api/pipelines/overview.mdx

@@ -98,15 +98,13 @@ logic including pre-processing, an unrolled diffusion loop, and post-processing
 
 ```python
 # make sure you're logged in with `huggingface-cli login`
-from torch import autocast
 from diffusers import StableDiffusionPipeline, LMSDiscreteScheduler
 
 pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4")
 pipe = pipe.to("cuda")
 
 prompt = "a photo of an astronaut riding a horse on mars"
-with autocast("cuda"):
-    image = pipe(prompt).images[0]
+image = pipe(prompt).images[0]
 
 image.save("astronaut_rides_horse.png")
 ```
@@ -116,7 +114,6 @@ image.save("astronaut_rides_horse.png")
 The `StableDiffusionImg2ImgPipeline` lets you pass a text prompt and an initial image to condition the generation of new images.
 
 ```python
-from torch import autocast
 import requests
 from PIL import Image
 from io import BytesIO
@@ -138,8 +135,7 @@ init_image = init_image.resize((768, 512))
 
 prompt = "A fantasy landscape, trending on artstation"
 
-with autocast("cuda"):
-    images = pipe(prompt=prompt, init_image=init_image, strength=0.75, guidance_scale=7.5).images
+images = pipe(prompt=prompt, init_image=init_image, strength=0.75, guidance_scale=7.5).images
 
 images[0].save("fantasy_landscape.png")
 ```
@@ -157,7 +153,6 @@ The `StableDiffusionInpaintPipeline` lets you edit specific parts of an image by
 ```python
 from io import BytesIO
 
-from torch import autocast
 import requests
 import PIL
 
@@ -181,8 +176,7 @@ pipe = StableDiffusionInpaintPipeline.from_pretrained(
 ).to(device)
 
 prompt = "a cat sitting on a bench"
-with autocast("cuda"):
-    images = pipe(prompt=prompt, init_image=init_image, mask_image=mask_image, strength=0.75).images
+images = pipe(prompt=prompt, init_image=init_image, mask_image=mask_image, strength=0.75).images
 
 images[0].save("cat_on_bench.png")
 ```

docs/source/optimization/fp16.mdx

@@ -68,7 +68,7 @@ Despite the precision loss, in our experience the final image results look the s
 
 ## Half precision weights
 
-To save more GPU memory, you can load the model weights directly in half precision. This involves loading the float16 version of the weights, which was saved to a branch named `fp16`, and telling PyTorch to use the `float16` type when loading them:
+To save more GPU memory and get even more speed, you can load and run the model weights directly in half precision. This involves loading the float16 version of the weights, which was saved to a branch named `fp16`, and telling PyTorch to use the `float16` type when loading them:
 
 ```Python
 pipe = StableDiffusionPipeline.from_pretrained(
@@ -76,6 +76,10 @@ pipe = StableDiffusionPipeline.from_pretrained(
     revision="fp16",
     torch_dtype=torch.float16,
 )
+pipe = pipe.to("cuda")
+
+prompt = "a photo of an astronaut riding a horse on mars"
+image = pipe(prompt).images[0]  
 ```
 
 ## Sliced attention for additional memory savings
@@ -101,8 +105,7 @@ pipe = pipe.to("cuda")
 
 prompt = "a photo of an astronaut riding a horse on mars"
 pipe.enable_attention_slicing()
-with torch.autocast("cuda"):
-    image = pipe(prompt).images[0]  
+image = pipe(prompt).images[0]  
 ```
 
 There's a small performance penalty of about 10% slower inference times, but this method allows you to use Stable Diffusion in as little as 3.2 GB of VRAM!

docs/source/training/text_inversion.mdx

@@ -109,7 +109,6 @@ A full training run takes ~1 hour on one V100 GPU.
 Once you have trained a model using above command, the inference can be done simply using the `StableDiffusionPipeline`. Make sure to include the `placeholder_token` in your prompt.
 
 ```python
-from torch import autocast
 from diffusers import StableDiffusionPipeline
 
 model_id = "path-to-your-trained-model"
@@ -117,8 +116,7 @@ pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float
 
 prompt = "A <cat-toy> backpack"
 
-with autocast("cuda"):
-    image = pipe(prompt, num_inference_steps=50, guidance_scale=7.5).images[0]
+image = pipe(prompt, num_inference_steps=50, guidance_scale=7.5).images[0]
 
 image.save("cat-backpack.png")
 ```

docs/source/using-diffusers/img2img.mdx

@@ -15,7 +15,6 @@ specific language governing permissions and limitations under the License.
 The [`StableDiffusionImg2ImgPipeline`] lets you pass a text prompt and an initial image to condition the generation of new images.
 
 ```python
-from torch import autocast
 import requests
 from PIL import Image
 from io import BytesIO
@@ -37,8 +36,7 @@ init_image = init_image.resize((768, 512))
 
 prompt = "A fantasy landscape, trending on artstation"
 
-with autocast("cuda"):
-    images = pipe(prompt=prompt, init_image=init_image, strength=0.75, guidance_scale=7.5).images
+images = pipe(prompt=prompt, init_image=init_image, strength=0.75, guidance_scale=7.5).images
 
 images[0].save("fantasy_landscape.png")
 ```

docs/source/using-diffusers/inpaint.mdx

@@ -17,7 +17,6 @@ The [`StableDiffusionInpaintPipeline`] lets you edit specific parts of an image
 ```python
 from io import BytesIO
 
-from torch import autocast
 import requests
 import PIL
 
@@ -41,8 +40,7 @@ pipe = StableDiffusionInpaintPipeline.from_pretrained(
 ).to(device)
 
 prompt = "a cat sitting on a bench"
-with autocast("cuda"):
-    images = pipe(prompt=prompt, init_image=init_image, mask_image=mask_image, strength=0.75).images
+images = pipe(prompt=prompt, init_image=init_image, mask_image=mask_image, strength=0.75).images
 
 images[0].save("cat_on_bench.png")
 ```

examples/dreambooth/README.md

@@ -125,8 +125,6 @@ accelerate launch train_dreambooth.py \
 Once you have trained a model using above command, the inference can be done simply using the `StableDiffusionPipeline`. Make sure to include the `identifier`(e.g. sks in above example) in your prompt.
 
 ```python
-
-from torch import autocast
 from diffusers import StableDiffusionPipeline
 import torch
 
@@ -134,9 +132,7 @@ model_id = "path-to-your-trained-model"
 pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16).to("cuda")
 
 prompt = "A photo of sks dog in a bucket"
-
-with autocast("cuda"):
-    image = pipe(prompt, num_inference_steps=50, guidance_scale=7.5).images[0]
+image = pipe(prompt, num_inference_steps=50, guidance_scale=7.5).images[0]
 
 image.save("dog-bucket.png")
 ```

examples/dreambooth/train_dreambooth.py

 import argparse
 import math
 import os
-from contextlib import nullcontext
 from pathlib import Path
 from typing import Optional
 
@@ -346,12 +345,10 @@ def main():
             sample_dataloader = accelerator.prepare(sample_dataloader)
             pipeline.to(accelerator.device)
 
-            context = torch.autocast("cuda") if accelerator.device.type == "cuda" else nullcontext
             for example in tqdm(
                 sample_dataloader, desc="Generating class images", disable=not accelerator.is_local_main_process
             ):
-                with context:
-                    images = pipeline(example["prompt"]).images
+                images = pipeline(example["prompt"]).images
 
                 for i, image in enumerate(images):
                     image.save(class_images_dir / f"{example['index'][i] + cur_class_images}.jpg")

examples/textual_inversion/README.md

@@ -74,8 +74,6 @@ A full training run takes ~1 hour on one V100 GPU.
 Once you have trained a model using above command, the inference can be done simply using the `StableDiffusionPipeline`. Make sure to include the `placeholder_token` in your prompt.
 
 ```python
-
-from torch import autocast
 from diffusers import StableDiffusionPipeline
 
 model_id = "path-to-your-trained-model"
@@ -83,8 +81,7 @@ pipe = StableDiffusionPipeline.from_pretrained(model_id,torch_dtype=torch.float1
 
 prompt = "A <cat-toy> backpack"
 
-with autocast("cuda"):
-    image = pipe(prompt, num_inference_steps=50, guidance_scale=7.5).images[0]
+image = pipe(prompt, num_inference_steps=50, guidance_scale=7.5).images[0]
 
 image.save("cat-backpack.png")
 ```