Diffusers documentation

Latent Consistency Distillation

Hugging Face's logo
Join the Hugging Face community

and get access to the augmented documentation experience

to get started

Latent Consistency Distillation

Latent Consistency Models (LCMs) are able to generate high-quality images in just a few steps, representing a big leap forward because many pipelines require at least 25+ steps. LCMs are produced by applying the latent consistency distillation method to any Stable Diffusion model. This method works by applying one-stage guided distillation to the latent space, and incorporating a skipping-step method to consistently skip timesteps to accelerate the distillation process (refer to section 4.1, 4.2, and 4.3 of the paper for more details).

If you’re training on a GPU with limited vRAM, try enabling gradient_checkpointing, gradient_accumulation_steps, and mixed_precision to reduce memory-usage and speedup training. You can reduce your memory-usage even more by enabling memory-efficient attention with xFormers and bitsandbytes’ 8-bit optimizer.

This guide will explore the train_lcm_distill_sd_wds.py script to help you become more familiar with it, and how you can adapt it for your own use-case.

Before running the script, make sure you install the library from source:

git clone https://github.com/huggingface/diffusers
cd diffusers
pip install .

Then navigate to the example folder containing the training script and install the required dependencies for the script you’re using:

cd examples/consistency_distillation
pip install -r requirements.txt

🤗 Accelerate is a library for helping you train on multiple GPUs/TPUs or with mixed-precision. It’ll automatically configure your training setup based on your hardware and environment. Take a look at the 🤗 Accelerate Quick tour to learn more.

Initialize an 🤗 Accelerate environment (try enabling torch.compile to significantly speedup training):

accelerate config

To setup a default 🤗 Accelerate environment without choosing any configurations:

accelerate config default

Or if your environment doesn’t support an interactive shell, like a notebook, you can use:

from accelerate.utils import write_basic_config

write_basic_config()

Lastly, if you want to train a model on your own dataset, take a look at the Create a dataset for training guide to learn how to create a dataset that works with the training script.

Script parameters

The following sections highlight parts of the training script that are important for understanding how to modify it, but it doesn’t cover every aspect of the script in detail. If you’re interested in learning more, feel free to read through the script and let us know if you have any questions or concerns.

The training script provides many parameters to help you customize your training run. All of the parameters and their descriptions are found in the parse_args() function. This function provides default values for each parameter, such as the training batch size and learning rate, but you can also set your own values in the training command if you’d like.

For example, to speedup training with mixed precision using the fp16 format, add the --mixed_precision parameter to the training command:

accelerate launch train_lcm_distill_sd_wds.py \
  --mixed_precision="fp16"

Most of the parameters are identical to the parameters in the Text-to-image training guide, so you’ll focus on the parameters that are relevant to latent consistency distillation in this guide.

  • --pretrained_teacher_model: the path to a pretrained latent diffusion model to use as the teacher model
  • --pretrained_vae_model_name_or_path: path to a pretrained VAE; the SDXL VAE is known to suffer from numerical instability, so this parameter allows you to specify an alternative VAE (like this VAE by madebyollin which works in fp16)
  • --w_min and --w_max: the minimum and maximum guidance scale values for guidance scale sampling
  • --num_ddim_timesteps: the number of timesteps for DDIM sampling
  • --loss_type: the type of loss (L2 or Huber) to calculate for latent consistency distillation; Huber loss is generally preferred because it’s more robust to outliers
  • --huber_c: the Huber loss parameter

Training script

The training script starts by creating a dataset class - Text2ImageDataset - for preprocessing the images and creating a training dataset.

def transform(example):
    image = example["image"]
    image = TF.resize(image, resolution, interpolation=transforms.InterpolationMode.BILINEAR)

    c_top, c_left, _, _ = transforms.RandomCrop.get_params(image, output_size=(resolution, resolution))
    image = TF.crop(image, c_top, c_left, resolution, resolution)
    image = TF.to_tensor(image)
    image = TF.normalize(image, [0.5], [0.5])

    example["image"] = image
    return example

For improved performance on reading and writing large datasets stored in the cloud, this script uses the WebDataset format to create a preprocessing pipeline to apply transforms and create a dataset and dataloader for training. Images are processed and fed to the training loop without having to download the full dataset first.

processing_pipeline = [
    wds.decode("pil", handler=wds.ignore_and_continue),
    wds.rename(image="jpg;png;jpeg;webp", text="text;txt;caption", handler=wds.warn_and_continue),
    wds.map(filter_keys({"image", "text"})),
    wds.map(transform),
    wds.to_tuple("image", "text"),
]

In the main() function, all the necessary components like the noise scheduler, tokenizers, text encoders, and VAE are loaded. The teacher UNet is also loaded here and then you can create a student UNet from the teacher UNet. The student UNet is updated by the optimizer during training.

teacher_unet = UNet2DConditionModel.from_pretrained(
    args.pretrained_teacher_model, subfolder="unet", revision=args.teacher_revision
)

unet = UNet2DConditionModel(**teacher_unet.config)
unet.load_state_dict(teacher_unet.state_dict(), strict=False)
unet.train()

Now you can create the optimizer to update the UNet parameters:

optimizer = optimizer_class(
    unet.parameters(),
    lr=args.learning_rate,
    betas=(args.adam_beta1, args.adam_beta2),
    weight_decay=args.adam_weight_decay,
    eps=args.adam_epsilon,
)

Create the dataset:

dataset = Text2ImageDataset(
    train_shards_path_or_url=args.train_shards_path_or_url,
    num_train_examples=args.max_train_samples,
    per_gpu_batch_size=args.train_batch_size,
    global_batch_size=args.train_batch_size * accelerator.num_processes,
    num_workers=args.dataloader_num_workers,
    resolution=args.resolution,
    shuffle_buffer_size=1000,
    pin_memory=True,
    persistent_workers=True,
)
train_dataloader = dataset.train_dataloader

Next, you’re ready to setup the training loop and implement the latent consistency distillation method (see Algorithm 1 in the paper for more details). This section of the script takes care of adding noise to the latents, sampling and creating a guidance scale embedding, and predicting the original image from the noise.

pred_x_0 = predicted_origin(
    noise_pred,
    start_timesteps,
    noisy_model_input,
    noise_scheduler.config.prediction_type,
    alpha_schedule,
    sigma_schedule,
)

model_pred = c_skip_start * noisy_model_input + c_out_start * pred_x_0

It gets the teacher model predictions and the LCM predictions next, calculates the loss, and then backpropagates it to the LCM.

if args.loss_type == "l2":
    loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
elif args.loss_type == "huber":
    loss = torch.mean(
        torch.sqrt((model_pred.float() - target.float()) ** 2 + args.huber_c**2) - args.huber_c
    )

If you want to learn more about how the training loop works, check out the Understanding pipelines, models and schedulers tutorial which breaks down the basic pattern of the denoising process.

Launch the script

Now you’re ready to launch the training script and start distilling!

For this guide, you’ll use the --train_shards_path_or_url to specify the path to the Conceptual Captions 12M dataset stored on the Hub here. Set the MODEL_DIR environment variable to the name of the teacher model and OUTPUT_DIR to where you want to save the model.

export MODEL_DIR="stable-diffusion-v1-5/stable-diffusion-v1-5"
export OUTPUT_DIR="path/to/saved/model"

accelerate launch train_lcm_distill_sd_wds.py \
    --pretrained_teacher_model=$MODEL_DIR \
    --output_dir=$OUTPUT_DIR \
    --mixed_precision=fp16 \
    --resolution=512 \
    --learning_rate=1e-6 --loss_type="huber" --ema_decay=0.95 --adam_weight_decay=0.0 \
    --max_train_steps=1000 \
    --max_train_samples=4000000 \
    --dataloader_num_workers=8 \
    --train_shards_path_or_url="pipe:curl -L -s https://huggingface.co/datasets/laion/conceptual-captions-12m-webdataset/resolve/main/data/{00000..01099}.tar?download=true" \
    --validation_steps=200 \
    --checkpointing_steps=200 --checkpoints_total_limit=10 \
    --train_batch_size=12 \
    --gradient_checkpointing --enable_xformers_memory_efficient_attention \
    --gradient_accumulation_steps=1 \
    --use_8bit_adam \
    --resume_from_checkpoint=latest \
    --report_to=wandb \
    --seed=453645634 \
    --push_to_hub

Once training is complete, you can use your new LCM for inference.

from diffusers import UNet2DConditionModel, DiffusionPipeline, LCMScheduler
import torch

unet = UNet2DConditionModel.from_pretrained("your-username/your-model", torch_dtype=torch.float16, variant="fp16")
pipeline = DiffusionPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", unet=unet, torch_dtype=torch.float16, variant="fp16")

pipeline.scheduler = LCMScheduler.from_config(pipe.scheduler.config)
pipeline.to("cuda")

prompt = "sushi rolls in the form of panda heads, sushi platter"

image = pipeline(prompt, num_inference_steps=4, guidance_scale=1.0).images[0]

LoRA

LoRA is a training technique for significantly reducing the number of trainable parameters. As a result, training is faster and it is easier to store the resulting weights because they are a lot smaller (~100MBs). Use the train_lcm_distill_lora_sd_wds.py or train_lcm_distill_lora_sdxl.wds.py script to train with LoRA.

The LoRA training script is discussed in more detail in the LoRA training guide.

Stable Diffusion XL

Stable Diffusion XL (SDXL) is a powerful text-to-image model that generates high-resolution images, and it adds a second text-encoder to its architecture. Use the train_lcm_distill_sdxl_wds.py script to train a SDXL model with LoRA.

The SDXL training script is discussed in more detail in the SDXL training guide.

Next steps

Congratulations on distilling a LCM model! To learn more about LCM, the following may be helpful:

< > Update on GitHub