Diffusers documentation

Working with big models

Hugging Face's logo
Join the Hugging Face community

and get access to the augmented documentation experience

to get started

Working with big models

A modern diffusion model, like Stable Diffusion XL (SDXL), is not just a single model, but a collection of multiple models. SDXL has four different model-level components:

  • A variational autoencoder (VAE)
  • Two text encoders
  • A UNet for denoising

Usually, the text encoders and the denoiser are much larger compared to the VAE.

As models get bigger and better, it’s possible your model is so big that even a single copy won’t fit in memory. But that doesn’t mean it can’t be loaded. If you have more than one GPU, there is more memory available to store your model. In this case, it’s better to split your model checkpoint into several smaller checkpoint shards.

When a text encoder checkpoint has multiple shards, like T5-xxl for SD3, it is automatically handled by the Transformers library as it is a required dependency of Diffusers when using the StableDiffusion3Pipeline. More specifically, Transformers will automatically handle the loading of multiple shards within the requested model class and get it ready so that inference can be performed.

The denoiser checkpoint can also have multiple shards and supports inference thanks to the Accelerate library.

Refer to the Handling big models for inference guide for general guidance when working with big models that are hard to fit into memory.

For example, let’s save a sharded checkpoint for the SDXL UNet:

from diffusers import UNet2DConditionModel

unet = UNet2DConditionModel.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0", subfolder="unet"
)
unet.save_pretrained("sdxl-unet-sharded", max_shard_size="5GB")

The size of the fp32 variant of the SDXL UNet checkpoint is ~10.4GB. Set the max_shard_size parameter to 5GB to create 3 shards. After saving, you can load them in StableDiffusionXLPipeline:

from diffusers import UNet2DConditionModel, StableDiffusionXLPipeline
import torch

unet = UNet2DConditionModel.from_pretrained(
    "sayakpaul/sdxl-unet-sharded", torch_dtype=torch.float16
)
pipeline = StableDiffusionXLPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0", unet=unet, torch_dtype=torch.float16
).to("cuda")

image = pipeline("a cute dog running on the grass", num_inference_steps=30).images[0]
image.save("dog.png")

If placing all the model-level components on the GPU at once is not feasible, use enable_model_cpu_offload() to help you:

- pipeline.to("cuda")
+ pipeline.enable_model_cpu_offload()

In general, we recommend sharding when a checkpoint is more than 5GB (in fp32).

Device placement

On distributed setups, you can run inference across multiple GPUs with Accelerate.

This feature is experimental and its APIs might change in the future.

With Accelerate, you can use the device_map to determine how to distribute the models of a pipeline across multiple devices. This is useful in situations where you have more than one GPU.

For example, if you have two 8GB GPUs, then using enable_model_cpu_offload() may not work so well because:

  • it only works on a single GPU
  • a single model might not fit on a single GPU (enable_sequential_cpu_offload() might work but it will be extremely slow and it is also limited to a single GPU)

To make use of both GPUs, you can use the “balanced” device placement strategy which splits the models across all available GPUs.

Only the “balanced” strategy is supported at the moment, and we plan to support additional mapping strategies in the future.

from diffusers import DiffusionPipeline
import torch

pipeline = DiffusionPipeline.from_pretrained(
-    "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, use_safetensors=True,
+    "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, use_safetensors=True, device_map="balanced"
)
image = pipeline("a dog").images[0]
image

You can also pass a dictionary to enforce the maximum GPU memory that can be used on each device:

from diffusers import DiffusionPipeline
import torch

max_memory = {0:"1GB", 1:"1GB"}
pipeline = DiffusionPipeline.from_pretrained(
    "runwayml/stable-diffusion-v1-5",
    torch_dtype=torch.float16,
    use_safetensors=True,
    device_map="balanced",
+   max_memory=max_memory
)
image = pipeline("a dog").images[0]
image

If a device is not present in max_memory, then it will be completely ignored and will not participate in the device placement.

By default, Diffusers uses the maximum memory of all devices. If the models don’t fit on the GPUs, they are offloaded to the CPU. If the CPU doesn’t have enough memory, then you might see an error. In that case, you could defer to using enable_sequential_cpu_offload() and enable_model_cpu_offload().

Call reset_device_map() to reset the device_map of a pipeline. This is also necessary if you want to use methods like to(), enable_sequential_cpu_offload(), and enable_model_cpu_offload() on a pipeline that was device-mapped.

pipeline.reset_device_map()

Once a pipeline has been device-mapped, you can also access its device map via hf_device_map:

print(pipeline.hf_device_map)

An example device map would look like so:

{'unet': 1, 'vae': 1, 'safety_checker': 0, 'text_encoder': 0}
< > Update on GitHub