diffusers/docs/source/optimization/fp16.mdx

<!--Copyright 2022 The HuggingFace Team. All rights reserved.

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
specific language governing permissions and limitations under the License.
-->

# Memory and speed

We present some techniques and ideas to optimize 🤗 Diffusers _inference_ for memory or speed.

## CUDA `autocast`

If you use a CUDA GPU, you can take advantage of `torch.autocast` to perform inference roughly twice as fast at the cost of slightly lower precision. All you need to do is put your inference call inside an `autocast` context manager. The following example shows how to do it using Stable Diffusion text-to-image generation as an example:

```Python
from torch import autocast
from diffusers import StableDiffusionPipeline

pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", use_auth_token=True)
pipe = pipe.to("cuda")

prompt = "a photo of an astronaut riding a horse on mars"
with autocast("cuda"):
    image = pipe(prompt).images[0]  
```

Despite the precision loss, in our experience the final image results look the same as the `float32` versions. Feel free to experiment and report back!

## 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:

```Python
pipe = StableDiffusionPipeline.from_pretrained(
    "CompVis/stable-diffusion-v1-4",
    revision="fp16",
    torch_dtype=torch.float16,
    use_auth_token=True
)
```

## Sliced attention for additional memory savings

For even additional memory savings, you can use a sliced version of attention that performs the computation in steps instead of all at once. 

<Tip>
Attention slicing is useful even if a batch size of just 1 is used - as long as the model uses more than one attention head. If there is more than one attention head the *QK^T* attention matrix can be computed sequentially for each head which can save a significant amount of memory.
</Tip>

To perform the attention computation sequentially over each head, you only need to invoke [`~StableDiffusionPipeline.enable_attention_slicing`] in your pipeline before inference, like here:

```Python
import torch
from diffusers import StableDiffusionPipeline

pipe = StableDiffusionPipeline.from_pretrained(
    "CompVis/stable-diffusion-v1-4",
    revision="fp16",
    torch_dtype=torch.float16,
    use_auth_token=True
)
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]  
```

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!