Diffusion Model Quantization

Ask about this topic

Overview

Diffusion model quantization studies how to use fewer bits for model weights, activations, and attention caches while keeping iterative generation close to full precision, enabling larger models to run with less memory and time. Common approaches include post-training quantization and quantization-aware training, with rotations, smoothing, outlier handling, low-rank compensation, and mixed precision used to control errors. Methods increasingly account for changing sensitivity across denoising steps, classifier-free guidance, and temporal consistency, while sampling-side corrections and fused low-bit kernels address error propagation and real deployment speed. W4A4 and even lower-bit settings can retain useful quality on image and video benchmarks such as COCO and VBench, but results remain model- and hardware-dependent: aggressive compression can harm fidelity, and speedups require kernels that use a device’s native low-precision operations.

Latest papers

CardsList

39 papers4 papers in the last four weeks, down 20% on the four weeks before. 0.0% of all new papers.

  1. PermuQuant: Lowering Per-Group Quantization Error by Reordering Channels for Diffusion Models

    May 10, 2026Yongsen Cheng, Kai Liu, Kaiwen Tao +5Diffusion Model QuantizationPost-Training Quantization

    Shanghai Jiao Tong University · Huawei Noah’s Ark Lab

  2. On the Quantization Robustness of Diffusion Language Models in Coding Benchmarks

    Apr 22, 2026Aarav Gupta, Gururaj Deshpande, Chandreyi ChakrabortyMixed-Precision QuantizationDiffusion Model Quantization

    College of Computing · Georgia Institute of Technology

  3. LoRaQ: Optimized Low Rank Approximation for 4-bit Quantization

    Apr 20, 2026Yann Bouquet, Alireza Khodamoradi, Sophie Yáng Shen +2Mixed-Precision QuantizationDiffusion Transformer

    Research and Advancement Development Team, Advanced Micro Devices, Inc, USA · Computer Vision Laboratory, Ecole Polytechnique F´ed´erale de Lausanne, Lausanne, Switzerland.

  4. PersonalQ: Select, Quantize, and Serve Personalized Diffusion Models for Efficient Inference

    Mar 24, 2026Qirui Wang, Qi Guo, Yiding Sun +4Diffusion Model ServingT2I Diffusion Models

    School of Software Engineering, Xi’an Jiaotong University, China · Nankai University, China

  5. 6Bit-Diffusion: Inference-Time Mixed-Precision Quantization for Video Diffusion Models

    Mar 19, 2026Rundong Su, Jintao Zhang, Zhihang Yuan +3Diffusion Model QuantizationMixed-Precision Quantization

    Fudan University · Tsinghua University · Shanghai Jiao Tong University

  6. Q-Drift: Quantization-Aware Drift Correction for Diffusion Model Sampling

    Mar 18, 2026Sooyoung Ryu, Mathieu Salzmann, Saqib JavedDiffusion Model QuantizationDiffusion Model Sampling

    Department of Computer Science and Engineering, Seoul National University, Seoul, South Korea · School of Computer and Communication Sciences, EPFL, Lausanne, Switzerland · Meta Reality Labs, Redmond, USA

  7. TMPDiff: Temporal Mixed-Precision for Diffusion Models

    Mar 14, 2026Basile Lewandowski, Simon Kurz, Aditya Shankar +3Diffusion Model QuantizationDiffusion Model Inference Acceleration

    University of Neuchâtel, Switzerland · TU Dortmund University, Germany · LAMARR Institute, Germany +2

  8. SemanticDialect: Semantic-Aware Mixed-Format Quantization for Video Diffusion Transformers

    Mar 3, 2026Wonsuk Jang, Thierry TambeDiffusion Model QuantizationNeural Network Quantization

    Stanford University

  9. LSGQuant: Layer-Sensitivity Guided Quantization for One-Step Diffusion Real-World Video Super-Resolution

    Feb 3, 2026Tianxing Wu, Zheng Chen, Cirou Xu +5Diffusion Model QuantizationVideo Super-Resolution

    Shanghai Jiao Tong University · Huawei