GPUSQ ViT

GPUSQ-ViT research focuses on efficiently deploying Vision Transformers (ViTs) on GPUs by employing quantization techniques to reduce computational cost and memory footprint without significant accuracy loss. Current efforts concentrate on developing novel quantization methods tailored to the unique activation distributions within ViTs, often involving mixed-precision strategies and addressing outliers to improve performance, particularly at low bit-widths (e.g., 4-bit). This work is significant because it enables the practical deployment of powerful ViT models on resource-constrained devices, expanding their applicability in various computer vision tasks.

Papers

August 6, 2024

DopQ-ViT: Towards Distribution-Friendly and Outlier-Aware Post-Training Quantization for Vision Transformers
Lianwei Yang, Haisong Gong, Qingyi Gu
Vision Transformer Layer Wise Quantization Quantization Model GPUSQ ViT

July 3, 2024

ADFQ-ViT: Activation-Distribution-Friendly Post-Training Quantization for Vision Transformers
Yanfeng Jiang, Ning Sun, Xueshuo Xie, Fei Yang, Tao Li
Vision Transformer Quantization Operator Uniform Quantization GPUSQ ViT

January 26, 2024

MPTQ-ViT: Mixed-Precision Post-Training Quantization for Vision Transformer
Yu-Shan Tai, An-Yeu, Wu
Vision Transformer Quantization Operator Low Bit Quantization GPUSQ ViT

May 18, 2023

Boost Vision Transformer with GPU-Friendly Sparsity and Quantization
Chong Yu, Tao Chen, Zhongxue Gan, Jiayuan Fan
Vision Transformer Quantization Operator Structured Sparsity Quantization Aware Knowledge Distillation GPUSQ ViT

December 16, 2022

RepQ-ViT: Scale Reparameterization for Post-Training Quantization of Vision Transformers
Zhikai Li, Junrui Xiao, Lianwei Yang, Qingyi Gu
Vision Transformer Quantization Operator Post Training Quantization Accurate Quantization GPUSQ ViT

September 13, 2022

PSAQ-ViT V2: Towards Accurate and General Data-Free Quantization for Vision Transformers
Zhikai Li, Mengjuan Chen, Junrui Xiao, Qingyi Gu
Vision Transformer Quantization Technique Data Free Quantization GPUSQ ViT

GPUSQ ViT

Papers

DopQ-ViT: Towards Distribution-Friendly and Outlier-Aware Post-Training Quantization for Vision Transformers

ADFQ-ViT: Activation-Distribution-Friendly Post-Training Quantization for Vision Transformers

MPTQ-ViT: Mixed-Precision Post-Training Quantization for Vision Transformer

Boost Vision Transformer with GPU-Friendly Sparsity and Quantization

RepQ-ViT: Scale Reparameterization for Post-Training Quantization of Vision Transformers

PSAQ-ViT V2: Towards Accurate and General Data-Free Quantization for Vision Transformers