Diffusers

Speed up inference

There are several ways to optimize 🤗 Diffusers for inference speed. As a general rule of thumb, we recommend using either xFormers or torch.nn.functional.scaled_dot_product_attention in PyTorch 2.0 for their memory-efficient attention.

In many cases, optimizing for speed or memory leads to improved performance in the other, so you should try to optimize for both whenever you can. This guide focuses on inference speed, but you can learn more about preserving memory in the Reduce memory usage guide.

The results below are obtained from generating a single 512x512 image from the prompt a photo of an astronaut riding a horse on mars with 50 DDIM steps on a Nvidia Titan RTX, demonstrating the speed-up you can expect.

	latency	speed-up
original	9.50s	x1
fp16	3.61s	x2.63
channels last	3.30s	x2.88
traced UNet	3.21s	x2.96
memory efficient attention	2.63s	x3.61

Use TensorFloat-32

On Ampere and later CUDA devices, matrix multiplications and convolutions can use the TensorFloat-32 (TF32) mode for faster, but slightly less accurate computations. By default, PyTorch enables TF32 mode for convolutions but not matrix multiplications. Unless your network requires full float32 precision, we recommend enabling TF32 for matrix multiplications. It can significantly speeds up computations with typically negligible loss in numerical accuracy.

import torch

torch.backends.cuda.matmul.allow_tf32 = True

You can learn more about TF32 in the Mixed precision training guide.

Half-precision weights

To save GPU memory and get more speed, try loading and running the model weights directly in half-precision or float16:

import torch
from diffusers import DiffusionPipeline

pipe = DiffusionPipeline.from_pretrained(
    "runwayml/stable-diffusion-v1-5",
    torch_dtype=torch.float16,
    use_safetensors=True,
)
pipe = pipe.to("cuda")

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

Don’t use torch.autocast in any of the pipelines as it can lead to black images and is always slower than pure float16 precision.

Distilled model

You could also use a distilled Stable Diffusion model and autoencoder to speed up inference. During distillation, many of the UNet’s residual and attention blocks are shed to reduce the model size. The distilled model is faster and uses less memory while generating images of comparable quality to the full Stable Diffusion model.

Learn more about in the Distilled Stable Diffusion inference guide!