Nanoscaling Floating-Point (NxFP): NanoMantissa, Adaptive Microexponents, and Code Recycling for Direct-Cast Compression of Large Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Lo, Yun-Chen, Wei, Gu-Yeon, Brooks, David
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910765043154944
author Lo, Yun-Chen
Wei, Gu-Yeon
Brooks, David
author_facet Lo, Yun-Chen
Wei, Gu-Yeon
Brooks, David
contents As cutting-edge large language models (LLMs) continue to transform various industries, their fast-growing model size and sequence length have led to memory traffic and capacity challenges. Recently, AMD, Arm, Intel, Meta, Microsoft, NVIDIA, and Qualcomm have proposed a Microscaling standard (Mx), which augments block floating-point with microexponents to achieve promising perplexity-to-footprint trade-offs. However, the Microscaling suffers from significant perplexity degradation on modern LLMs with less than six bits. This paper profiles modern LLMs and identifies three main challenges of low-bit Microscaling format, i.e., inaccurate tracking of outliers, vacant quantization levels, and wasted binary code. In response, Nanoscaling (NxFP) proposes three techniques, i.e., NanoMantissa, Adaptive Microexponent, and Code Recycling to enable better accuracy and smaller memory footprint than state-of-the-art MxFP. Experimental results on direct-cast inference across various modern LLMs demonstrate that our proposed methods outperform state-of-the-art MxFP by up to 0.64 in perplexity and by up to 30% in accuracy on MMLU benchmarks. Furthermore, NxFP reduces memory footprint by up to 16% while achieving comparable perplexity as MxFP.
format Preprint
id arxiv_https___arxiv_org_abs_2412_19821
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Nanoscaling Floating-Point (NxFP): NanoMantissa, Adaptive Microexponents, and Code Recycling for Direct-Cast Compression of Large Language Models
Lo, Yun-Chen
Wei, Gu-Yeon
Brooks, David
Hardware Architecture
Artificial Intelligence
Distributed, Parallel, and Cluster Computing
Machine Learning
I.2.7; E.4
As cutting-edge large language models (LLMs) continue to transform various industries, their fast-growing model size and sequence length have led to memory traffic and capacity challenges. Recently, AMD, Arm, Intel, Meta, Microsoft, NVIDIA, and Qualcomm have proposed a Microscaling standard (Mx), which augments block floating-point with microexponents to achieve promising perplexity-to-footprint trade-offs. However, the Microscaling suffers from significant perplexity degradation on modern LLMs with less than six bits. This paper profiles modern LLMs and identifies three main challenges of low-bit Microscaling format, i.e., inaccurate tracking of outliers, vacant quantization levels, and wasted binary code. In response, Nanoscaling (NxFP) proposes three techniques, i.e., NanoMantissa, Adaptive Microexponent, and Code Recycling to enable better accuracy and smaller memory footprint than state-of-the-art MxFP. Experimental results on direct-cast inference across various modern LLMs demonstrate that our proposed methods outperform state-of-the-art MxFP by up to 0.64 in perplexity and by up to 30% in accuracy on MMLU benchmarks. Furthermore, NxFP reduces memory footprint by up to 16% while achieving comparable perplexity as MxFP.
title Nanoscaling Floating-Point (NxFP): NanoMantissa, Adaptive Microexponents, and Code Recycling for Direct-Cast Compression of Large Language Models
topic Hardware Architecture
Artificial Intelligence
Distributed, Parallel, and Cluster Computing
Machine Learning
I.2.7; E.4
url https://arxiv.org/abs/2412.19821