Saved in:
Bibliographic Details
Main Authors: Tseng, Albert, Chee, Jerry, Sun, Qingyao, Kuleshov, Volodymyr, De Sa, Christopher
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2402.04396
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909215701860352
author Tseng, Albert
Chee, Jerry
Sun, Qingyao
Kuleshov, Volodymyr
De Sa, Christopher
author_facet Tseng, Albert
Chee, Jerry
Sun, Qingyao
Kuleshov, Volodymyr
De Sa, Christopher
contents Post-training quantization (PTQ) reduces the memory footprint of LLMs by quantizing their weights to low-precision. In this work, we introduce QuIP#, a weight-only PTQ method that achieves state-of-the-art results in extreme compression regimes ($\le$ 4 bits per weight) using three novel techniques. First, QuIP# improves QuIP's (Chee et al., 2023) incoherence processing by using the randomized Hadamard transform, which is faster and has better theoretical properties. Second, QuIP# uses vector quantization to take advantage of the ball-shaped sub-Gaussian distribution that incoherent weights possess: specifically, we introduce a set of hardware-efficient codebooks based on the highly symmetric $E_8$ lattice, which achieves the optimal 8-dimension unit ball packing. Third, QuIP# uses fine-tuning to improve fidelity to the original model. Our experiments show that QuIP# outperforms existing PTQ methods, enables new behaviors in PTQ scaling, and supports fast inference. Our code can be found at https://github.com/Cornell-RelaxML/quip-sharp.
format Preprint
id arxiv_https___arxiv_org_abs_2402_04396
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle QuIP#: Even Better LLM Quantization with Hadamard Incoherence and Lattice Codebooks
Tseng, Albert
Chee, Jerry
Sun, Qingyao
Kuleshov, Volodymyr
De Sa, Christopher
Machine Learning
Artificial Intelligence
Computation and Language
Post-training quantization (PTQ) reduces the memory footprint of LLMs by quantizing their weights to low-precision. In this work, we introduce QuIP#, a weight-only PTQ method that achieves state-of-the-art results in extreme compression regimes ($\le$ 4 bits per weight) using three novel techniques. First, QuIP# improves QuIP's (Chee et al., 2023) incoherence processing by using the randomized Hadamard transform, which is faster and has better theoretical properties. Second, QuIP# uses vector quantization to take advantage of the ball-shaped sub-Gaussian distribution that incoherent weights possess: specifically, we introduce a set of hardware-efficient codebooks based on the highly symmetric $E_8$ lattice, which achieves the optimal 8-dimension unit ball packing. Third, QuIP# uses fine-tuning to improve fidelity to the original model. Our experiments show that QuIP# outperforms existing PTQ methods, enables new behaviors in PTQ scaling, and supports fast inference. Our code can be found at https://github.com/Cornell-RelaxML/quip-sharp.
title QuIP#: Even Better LLM Quantization with Hadamard Incoherence and Lattice Codebooks
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2402.04396