Saved in:
Bibliographic Details
Main Authors: Wang, Jinguang, Yin, Yuexi, Sun, Haifeng, Qi, Qi, Wang, Jingyu, Zhuang, Zirui, Yang, Tingting, Liao, Jianxin
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2406.18832
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929401628721152
author Wang, Jinguang
Yin, Yuexi
Sun, Haifeng
Qi, Qi
Wang, Jingyu
Zhuang, Zirui
Yang, Tingting
Liao, Jianxin
author_facet Wang, Jinguang
Yin, Yuexi
Sun, Haifeng
Qi, Qi
Wang, Jingyu
Zhuang, Zirui
Yang, Tingting
Liao, Jianxin
contents Quantizing the activations of large language models (LLMs) has been a significant challenge due to the presence of structured outliers. Most existing methods focus on the per-token or per-tensor quantization of activations, making it difficult to achieve both accuracy and hardware efficiency. To address this problem, we propose OutlierTune, an efficient per-channel post-training quantization (PTQ) method for the activations of LLMs. OutlierTune consists of two components: pre-execution of dequantization and symmetrization. The pre-execution of dequantization updates the model weights by the activation scaling factors, avoiding the internal scaling and costly additional computational overheads brought by the per-channel activation quantization. The symmetrization further reduces the quantization differences arising from the weight updates by ensuring the balanced numerical ranges across different activation channels. OutlierTune is easy to implement and hardware-efficient, introducing almost no additional computational overheads during the inference. Extensive experiments show that the proposed framework outperforms existing methods across multiple different tasks. Demonstrating better generalization, this framework improves the Int6 quantization of the instruction-tuning LLMs, such as OPT-IML, to the same level as half-precision (FP16). Moreover, we have shown that the proposed framework is 1.48x faster than the FP16 implementation while reducing approximately 2x memory usage.
format Preprint
id arxiv_https___arxiv_org_abs_2406_18832
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle OutlierTune: Efficient Channel-Wise Quantization for Large Language Models
Wang, Jinguang
Yin, Yuexi
Sun, Haifeng
Qi, Qi
Wang, Jingyu
Zhuang, Zirui
Yang, Tingting
Liao, Jianxin
Computation and Language
Quantizing the activations of large language models (LLMs) has been a significant challenge due to the presence of structured outliers. Most existing methods focus on the per-token or per-tensor quantization of activations, making it difficult to achieve both accuracy and hardware efficiency. To address this problem, we propose OutlierTune, an efficient per-channel post-training quantization (PTQ) method for the activations of LLMs. OutlierTune consists of two components: pre-execution of dequantization and symmetrization. The pre-execution of dequantization updates the model weights by the activation scaling factors, avoiding the internal scaling and costly additional computational overheads brought by the per-channel activation quantization. The symmetrization further reduces the quantization differences arising from the weight updates by ensuring the balanced numerical ranges across different activation channels. OutlierTune is easy to implement and hardware-efficient, introducing almost no additional computational overheads during the inference. Extensive experiments show that the proposed framework outperforms existing methods across multiple different tasks. Demonstrating better generalization, this framework improves the Int6 quantization of the instruction-tuning LLMs, such as OPT-IML, to the same level as half-precision (FP16). Moreover, we have shown that the proposed framework is 1.48x faster than the FP16 implementation while reducing approximately 2x memory usage.
title OutlierTune: Efficient Channel-Wise Quantization for Large Language Models
topic Computation and Language
url https://arxiv.org/abs/2406.18832