A Comprehensive Evaluation on Quantization Techniques for Large Language Models

TL;DR

This paper provides a comprehensive, fair comparison of recent quantization techniques for large language models, analyzing their components, settings, and data formats to guide future improvements.

Contribution

It introduces a unified evaluation framework by decoupling quantization methods into two steps and systematically compares various settings and data formats.

Findings

Optimized rotation and scaling improve pre-quantization performance.

Combining low-rank compensation with GPTQ can outperform GPTQ alone.

Finer granularity enhances performance but increases storage overhead.

Abstract

For large language models (LLMs), post-training quantization (PTQ) can significantly reduce memory footprint and computational overhead. Model quantization is rapidly evolving. Though many papers report breakthrough results, they are often evaluated under different settings because a method typically contains multiple components. Analyzing connections among existing methods is important for deeper understanding. To bridge these gaps, we conduct an extensive review of state-of-the-art methods and perform comprehensive evaluations under the same conditions for fair comparison. To our knowledge, such a fair and extensive investigation remains critically underexplored. To better understand connections, first, we decouple published quantization methods into two steps: pre-quantization transformation and quantization error mitigation. The former is a preprocessing step that reduces outlier impact by flattening the data distribution; the latter offsets quantization errors to improve performance. Second, we evaluate and analyze the impact of different settings, including granularity and symmetry. Third, we analyze and evaluate the latest MXFP4 and NVFP4 data formats and their performance. Our experiments first demonstrate that optimized rotation and scaling yield the best pre-quantization performance, and that combining low-rank compensation with GPTQ can occasionally outperform GPTQ alone for error mitigation. Second, finer granularity improves performance but increases storage overhead. Third, we find that scaling-factor format and precision greatly affect FP4 performance, and that rotation-based strategies effective for INT4 offer limited gains for MXFP4 and NVFP4, motivating further study.