🔍 Dive into the cutting-edge with this curated list of papers on Vision Transformers (ViT) quantization and hardware acceleration, featured in top-tier AI conferences and journals. This collection is meticulously organized and draws upon insights from our comprehensive survey:

[Arxiv] Model Quantization and Hardware Acceleration for Vision Transformers: A Comprehensive Survey

Table of Contents

• Accuracy Enhancement
  • Activation Quantization Optimization
  • Calibration Optimization For PTQ
  • Gradient-base Otimization For QAT
  • Binary Quantization
• Hardware Acceleration
  • Non-linear Operations Acceleration
  • Hardware Accelerator

Model Quantization

Activation Quantization Optimization

Date	Title	Paper	Code
2021.11	“PTQ4ViT: Post-training Quantization for Vision Transformers with Twin Uniform Quantization”	[ECCV‘22]	[code]
2021.11	“FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer”	[IJCAI’22]	[code]
2022.12	“RepQ-ViT: Scale Reparameterization for Post-Training Quantization of Vision Transformers”	[ICCV‘23]	[code]
2023.03	“Towards Accurate Post-Training Quantization for Vision Transformer”	[MM’22]	-
2023.05	“TSPTQ-ViT: Two-scaled post-training quantization for vision transformer”	[ICASSP‘23]	-
2023.11	“I&S-ViT: An Inclusive & Stable Method for Pushing the Limit of Post-Training ViTs Quantization”	[Arxiv]	[code]
2024.01	“MPTQ-ViT: Mixed-Precision Post-Training Quantization for Vision Transformer”	[Arxiv]	-
2024.01	“LRP-QViT: Mixed-Precision Vision Transformer Quantization via Layer-wise Relevance Propagation”	[Arxiv]	-
2024.02	“RepQuant: Towards Accurate Post-Training Quantization of Large Transformer Models via Scale Reparameterization”	[Arxiv]	-
2024.04	“Instance-Aware Group Quantization for Vision Transformers”	[Arxiv]	-
2024.05	“P^2-ViT: Power-of-Two Post-Training Quantization and Acceleration for Fully Quantized Vision Transformer”	[Arxiv]	[code]

Calibration Optimization For PTQ

Date	Title	Paper	Code
2021.06	“Post-Training Quantization for Vision Transformer”	[NIPS 2021]	[code]
2021.11	“PTQ4ViT: Post-training Quantization for Vision Transformers with Twin Uniform Quantization”	[ECCV’22]	[code]
2022.03	“Patch Similarity Aware Data-Free Quantization for Vision Transformers”	[ECCV‘22]	[code]
2022.09	“PSAQ-ViT V2: Towards Accurate and General Data-Free Quantization for Vision Transformers”	[TNNLS’23]	[code]
2022.11	“NoisyQuant: Noisy Bias-Enhanced Post-Training Activation Quantization for Vision Transformers”	[CVPR‘23]	-
2023.03	“Towards Accurate Post-Training Quantization for Vision Transformer”	[MM’22]	-
2023.05	“Finding Optimal Numerical Format for Sub-8-Bit Post-Training Quantization of Vision Transformers”	[ICASSP‘23]	-
2023.08	“Jumping through Local Minima: Quantization in the Loss Landscape of Vision Transformers”	[ICCV’23]	[code]
2023.10	“LLM-FP4: 4-Bit Floating-Point Quantized Transformers”	[EMNLP‘23]	[code]
2024.05	“P^2-ViT: Power-of-Two Post-Training Quantization and Acceleration for Fully Quantized Vision Transformer”	[Arxiv]	[code]

Gradient-base Optimization For QAT

Date	Title	Paper	Code
2022.01	“TerViT: An Efficient Ternary Vision Transformer”	[Arxiv]	-
2022.10	“Q-ViT: Accurate and Fully Quantized Low-bit Vision Transformer”	[NIPS’22]	[code]
2022.12	“Quantformer: Learning Extremely Low-Precision Vision Transformers”	[TPAMI‘22]	-
2023.02	“Oscillation-free Quantization for Low-bit Vision Transformers”	[PMLR’23]	[code]
2023.05	“Boost Vision Transformer with GPU-Friendly Sparsity and Quantization”	[CVPR‘23]	-
2023.06	“Bit-Shrinking: Limiting Instantaneous Sharpness for Improving Post-Training Quantization”	[CVPR’23]	-
2023.07	“Variation-aware Vision Transformer Quantization”	[Arxiv]	[code]
2023.12	“PackQViT: Faster Sub-8-bit Vision Transformers via Full and Packed Quantization on the Mobile”	[NIPS‘23]	-

Binary Quantization

Date	Title	Paper	Code
2022.11	“BiViT: Extremely Compressed Binary Vision Transformer”	[ICCV’23]	-
2023.05	“BinaryViT: Towards Efficient and Accurate Binary Vision Transformers”	[Arxiv]	-
2023.06	“BinaryViT: Pushing Binary Vision Transformers Towards Convolutional Models”	[CVPR‘23]	[code]
2024.05	“BinaryFormer: A Hierarchical-Adaptive Binary Vision Transformer (ViT) for Efficient Computing”	[TII]	-

Hardware Acceleration

Non-linear Operations Acceleration

Date	Title	Paper	Code
2021.11	“FQ-ViT: Post-Training Quantization for Fully Quantized Vision Transformer”	[IJCAI’22]	[code]
2022.07	“I-ViT: Integer-only Quantization for Efficient Vision Transformer Inference”	[ICCV‘23]	[code]
2023.06	“Practical Edge Kernels for Integer-Only Vision Transformers Under Post-training Quantization”	[MLSYS’23]	-
2023.10	“SOLE: Hardware-Software Co-design of Softmax and LayerNorm for Efficient Transformer Inference”	[ICCAD‘23]	-
2023.12	“PackQViT: Faster Sub-8-bit Vision Transformers via Full and Packed Quantization on the Mobile”	[NIPS’23]	-
2024.05	“P^2-ViT: Power-of-Two Post-Training Quantization and Acceleration for Fully Quantized Vision Transformer”	[Arxiv]	[code]

Hardware Accelerator

Date	Title	Paper	Code
2022.01	“VAQF: Fully Automatic Software-Hardware Co-Design Framework for Low-Bit Vision Transformer”	[Arxiv]	-
2022.08	“Auto-ViT-Acc: An FPGA-Aware Automatic Acceleration Framework for Vision Transformer with Mixed-Scheme Quantization”	[FPL‘22]	-
2023.10	“An Integer-Only and Group-Vector Systolic Accelerator for Efficiently Mapping Vision Transformer on Edge”	[TCAS-I’23]	-
2023.10	“SOLE: Hardware-Software Co-design of Softmax and LayerNorm for Efficient Transformer Inference”	[ICCAD‘23]	-
2024.05	“P^2-ViT: Power-of-Two Post-Training Quantization and Acceleration for Fully Quantized Vision Transformer”	[Arxiv]	[code]

Citation

If you find our survey useful or relevant to your research, please kindly cite our paper:

@misc{du2024model,
      title={Model Quantization and Hardware Acceleration for Vision Transformers: A Comprehensive Survey}, 
      author={Dayou Du and Gu Gong and Xiaowen Chu},
      year={2024},
      eprint={2405.00314},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}