QuEPT: Quantized Elastic Precision Transformers with One-Shot Calibration for Multi-Bit Switching

TL;DR

QuEPT introduces a novel post-training quantization method for Transformers that enables efficient multi-bit deployment with one-shot calibration, dynamic bit-width adaptation, and improved robustness through token merging and low-rank adapters.

Contribution

The paper presents QuEPT, a new elastic quantization scheme for Transformers that supports real-time switching between quantization precisions with minimal calibration, enhancing efficiency and robustness.

Findings

Achieves comparable or superior performance to state-of-the-art methods.

Supports real-time switching between uniform and mixed precision quantization.

Demonstrates effectiveness on large language models.

Abstract

Elastic precision quantization enables multi-bit deployment via a single optimization pass, fitting diverse quantization scenarios.Yet, the high storage and optimization costs associated with the Transformer architecture, research on elastic quantization remains limited, particularly for large language models.This paper proposes QuEPT, an efficient post-training scheme that reconstructs block-wise multi-bit errors with one-shot calibration on a small data slice. It can dynamically adapt to various predefined bit-widths by cascading different low-rank adapters, and supports real-time switching between uniform quantization and mixed precision quantization without repeated optimization. To enhance accuracy and robustness, we introduce Multi-Bit Token Merging (MB-ToMe) to dynamically fuse token features across different bit-widths, improving robustness during bit-width switching. Additionally, we propose Multi-Bit Cascaded Low-Rank adapters (MB-CLoRA) to strengthen correlations between bit-width groups, further improve the overall performance of QuEPT. Extensive experiments demonstrate that QuEPT achieves comparable or better performance to existing state-of-the-art post-training quantization methods.Our code is available at https://github.com/xuke225/QuEPT