LeanQuant: Accurate and Scalable Large Language Model Quantization with Loss-error-aware Grid

TL;DR

LeanQuant introduces a loss-error-aware grid-based quantization method for large language models, improving accuracy and scalability while maintaining compatibility with popular frameworks, enabling efficient quantization of models like Llama-3.1 405B.

Contribution

It proposes a novel loss-error-aware grid approach that overcomes limitations of prior min-max affine grids, enhancing model quality and framework compatibility.

Findings

Achieves high accuracy in quantizing large models like Llama-3.1 405B.

Demonstrates scalability by quantizing a 405B parameter model in 21 hours.

Outperforms existing methods in model quality and versatility.

Abstract

Large language models (LLMs) have shown immense potential across various domains, but their high memory requirements and inference costs remain critical challenges for deployment. Post-training quantization (PTQ) has emerged as a promising technique to reduce memory requirements and decoding latency. However, recent accurate quantization methods often depend on specialized computations or custom data formats to achieve better model quality, which limits their compatibility with popular frameworks, as they require dedicated inference kernels tailored to specific hardware and software platforms, hindering wider adoption. Furthermore, many competitive methods have high resource requirements and computational overhead for quantizing models, making it challenging to scale them to hundreds of billions of parameters. In response to these challenges, we propose LeanQuant (Loss-Error-Aware Network Quantization), a novel quantization method that is accurate, versatile, and scalable. In the existing popular iterative loss-error-based quantization framework, we identify a critical limitation in prior methods: the min-max affine quantization grid fails to preserve model quality due to outliers in inverse Hessian diagonals. To overcome this fundamental issue, we propose learning loss-error-aware grids, instead of using non-adaptive min-max affine grids. Our approach not only produces quantized models that are more accurate but also generalizes to a wider range of quantization types, including affine and non-uniform quantization, enhancing compatibility with more frameworks. Extensive experiments with recent LLMs demonstrate that LeanQuant is highly accurate, comparing favorably against competitive baselines in model quality, and scalable, achieving very accurate quantization of Llama-3.1 405B, one of the largest open-source LLMs to date, using two Quadro RTX 8000-48GB GPUs in 21 hours.