Latent Consistency Distillation Example:

Latent Consistency Models (LCMs) is a method to distill a latent diffusion model to enable swift inference with minimal steps. This example demonstrates how to use latent consistency distillation to distill stable-diffusion-v1.5 for inference with few timesteps.

Full model distillation

Running locally with PyTorch

Installing the dependencies

Before running the scripts, make sure to install the library's training dependencies:

Important

To make sure you can successfully run the latest versions of the example scripts, we highly recommend installing from source and keeping the install up to date as we update the example scripts frequently and install some example-specific requirements. To do this, execute the following steps in a new virtual environment:

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

Then cd in the example folder and run

pip install -r requirements.txt

And initialize an 🤗 Accelerate environment with:

accelerate config

Or for a default accelerate configuration without answering questions about your environment

accelerate config default

Or if your environment doesn't support an interactive shell e.g. a notebook

from accelerate.utils import write_basic_config
write_basic_config()

When running accelerate config, if we specify torch compile mode to True there can be dramatic speedups.

Example

The following uses the Conceptual Captions 12M (CC12M) dataset as an example, and for illustrative purposes only. For best results you may consider large and high-quality text-image datasets such as LAION. You may also need to search the hyperparameter space according to the dataset you use.

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

accelerate launch train_lcm_distill_sd_wds.py \
    --pretrained_teacher_model=$MODEL_NAME \
    --output_dir=$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

LCM-LoRA

Instead of fine-tuning the full model, we can also just train a LoRA that can be injected into any SDXL model.

Example

The following uses the Conceptual Captions 12M (CC12M) dataset as an example. For best results you may consider large and high-quality text-image datasets such as LAION.

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

accelerate launch train_lcm_distill_lora_sd_wds.py \
    --pretrained_teacher_model=$MODEL_NAME \
    --output_dir=$OUTPUT_DIR \
    --mixed_precision=fp16 \
    --resolution=512 \
    --lora_rank=64 \
    --learning_rate=1e-4 --loss_type="huber" --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 \