Bit Width Quantization

Bit width quantization aims to reduce the computational cost and memory footprint of deep learning models by representing their weights and activations using fewer bits, thereby enabling deployment on resource-constrained devices. Current research focuses on developing adaptive and mixed-precision quantization techniques, employing methods like layer-wise relevance propagation to optimize bit allocation across different layers or regions of a model, and incorporating reinforcement learning to dynamically adjust bit-widths based on data quality. These advancements are crucial for deploying large models like Vision Transformers and Large Language Models on edge devices and improving the efficiency of various applications, including image super-resolution and natural language processing.

Papers

April 4, 2024

AdaBM: On-the-Fly Adaptive Bit Mapping for Image Super-Resolution
Cheeun Hong, Kyoung Mu Lee
Image Super Resolution Adaptive Quantization Bit Allocation Bit Width Quantization

February 7, 2024

ApiQ: Finetuning of 2-Bit Quantized Large Language Model
Baohao Liao, Christian Herold, Shahram Khadivi, Christof Monz
Large Language Model Fine Tuning Quantization Technique Memory Efficient Fine Tuning Bit Width Quantization

January 20, 2024

LRP-QViT: Mixed-Precision Vision Transformer Quantization via Layer-wise Relevance Propagation
Navin Ranjan, Andreas Savakis
Vision Transformer Mixed Precision Quantization Layer Wise Relevance Propagation Vision Transformer Quantization Bit Width Quantization

July 6, 2023

Free Bits: Latency Optimization of Mixed-Precision Quantized Neural Networks on the Edge
Georg Rutishauser, Francesco Conti, Luca Benini
Neural Network Extreme Edge P Bit Mixed Precision Differentiable Search Latency Optimization Bit Width Quantization

May 14, 2023

MBQuant: A Novel Multi-Branch Topology Method for Arbitrary Bit-width Network Quantization
Yunshan Zhong, Yuyao Zhou, Fei Chao, Rongrong Ji
Topological Feature Low Bit Network Quantization Binary Activation Bit Width Quantization

February 9, 2023

Data Quality-aware Mixed-precision Quantization via Hybrid Reinforcement Learning
Yingchun Wang, Jingcai Guo, Song Guo, Weizhan Zhang
Mixed Precision Quantization Hybrid Reinforcement Learning Bit Width Quantization

July 21, 2022

CADyQ: Content-Aware Dynamic Quantization for Image Super-Resolution
Cheeun Hong, Sungyong Baik, Heewon Kim, Seungjun Nah, Kyoung Mu Lee
Image Super Resolution Dynamic Quantization Quantization Aware Bit Width Quantization

Bit Width Quantization

Papers

AdaBM: On-the-Fly Adaptive Bit Mapping for Image Super-Resolution

ApiQ: Finetuning of 2-Bit Quantized Large Language Model

LRP-QViT: Mixed-Precision Vision Transformer Quantization via Layer-wise Relevance Propagation

Free Bits: Latency Optimization of Mixed-Precision Quantized Neural Networks on the Edge

MBQuant: A Novel Multi-Branch Topology Method for Arbitrary Bit-width Network Quantization

Data Quality-aware Mixed-precision Quantization via Hybrid Reinforcement Learning

CADyQ: Content-Aware Dynamic Quantization for Image Super-Resolution