unconditional_image_generation

Training examples

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/junhsss/consistency-models
cd consistency-models
pip install .

Then cd in the example folder and run

pip install -r requirements.txt

Unconditional CIFAR-10

The command to train a UNet model on the CIFAR-10 dataset:

python train_unconditional.py \
  --dataset-name="cifar10" \
  --resolution=32 \
  --model-id="cm-cifar10-32" \
  --train-batch-size=32 \
  --max-epochs=100 \
  --use-ema \
  --learning-rate=1e-4 \
  --push-to-hub

In practice, you might want a larger batch size and a deeper architecture like NCSN++ or its variants. The proper configurations can be found in HF hub.