DiTAS: Quantizing Diffusion Transformers via Enhanced Activation Smoothing (WACV 2025)

[Paper] https://arxiv.org/abs/2409.07756

Zhenyuan Dong, Sai Qian Zhang

New York University

We propose DiTAS, a data-free post-training quantization (PTQ) method for efficient DiT inference. DiTAS relies on the proposed temporal-aggregated smoothing techniques to mitigate the impact of the channel-wise outliers within the input activations, leading to much lower quantization error under extremely low bitwidth. To further enhance the performance of the quantized DiT, we adopt the layer-wise grid search strategy to optimize the smoothing factor. Moreover, we integrate a training-free LoRA module for weight quantization, leveraging alternating optimization to minimize quantization errors without additional fine-tuning. Experimental results demonstrate that our approach enables 4-bit weight, 8-bit activation (W4A8) quantization for DiTs while maintaining comparable performance as the full-precision model.

Setup

First, download and set up the repo:

git clone https://github.com/DZY122/DiTAS.git
cd DiTAS

We provide an environment.yml file that can be used to create a Conda environment.

conda env create -f environment.yml
conda activate DiTAS

Then, choose and set the bit-width setting in config.py both for activation and weight quantization.

Quantizing Diffusion Transformers

DiTAS provides a quantization script for DiT in QuantDiT.py. This script can be used to quantize DiT models through advanced methods of DiTAS. The output is the quantized weight checkpoints and the optimized parameters for activation quantization:

python QuantDiT.py --image-size 256 --seed 1 --model DiT-XL/2 --act-bit 8 --weight-bit 4 --num-sampling-steps 100

Evaluation (FID, Inception Score, etc.)

DiTAS includes a sample_merge_TAS.py script which samples a large number of images from a DiTAS model in parallel. For the QKV and FC1 layers in DiT blocks, we merge the smoothing factor of activation into the side MLP. And we merge the smoothing factor of Projection layer’s activation into V’s weight. Finally, we operate on-the-fly activation smoothing for FC2 layer.

This script generates a folder of samples as well as a .npz file which can be directly used with ADM's TensorFlow evaluation suite to compute FID, Inception Score and other metrics. For example, to sample 10K images from our quantized DiT-XL/2 model over N GPUs, run:

torchrun --nnodes=1 --nproc_per_node=N sample_merge_TAS.py --image-size 256 --model DiT-XL/2 --num-fid-samples 10000 --act-bit 8 --weight-bit 4 --path /path/DiTAS_Model --num-sampling-steps 100

BibTeX

@InProceedings{Dong_2025_WACV,
    author    = {Dong, Zhenyuan and Zhang, Sai Qian},
    title     = {DiTAS: Quantizing Diffusion Transformers via Enhanced Activation Smoothing},
    booktitle = {Proceedings of the Winter Conference on Applications of Computer Vision (WACV)},
    month     = {February},
    year      = {2025},
    pages     = {4606-4615}
}