Accurate KV Cache Quantization with Outlier Tokens Tracing

TL;DR

This paper introduces a method to identify and exclude outlier tokens during KV cache quantization in large language models, significantly improving accuracy and reducing memory and computational costs.

Contribution

It proposes a novel outlier token tracing technique that enhances quantization accuracy by selectively excluding unusual tokens from quantization.

Findings

Achieves significant accuracy improvements with 2-bit quantization.

Reduces memory usage by 6.4 times.

Increases throughput by 2.3 times.

Abstract

The impressive capabilities of Large Language Models (LLMs) come at the cost of substantial computational resources during deployment. While KV Cache can significantly reduce recomputation during inference, it also introduces additional memory overhead. KV Cache quantization presents a promising solution, striking a good balance between memory usage and accuracy. Previous research has shown that the Keys are distributed by channel, while the Values are distributed by token. Consequently, the common practice is to apply channel-wise quantization to the Keys and token-wise quantization to the Values. However, our further investigation reveals that a small subset of unusual tokens exhibit unique characteristics that deviate from this pattern, which can substantially impact quantization accuracy. To address this, we develop a simple yet effective method to identify these tokens accurately during the decoding process and exclude them from quantization as outlier tokens, significantly improving overall accuracy. Extensive experiments show that our method achieves significant accuracy improvements under 2-bit quantization and can deliver a 6.4 times reduction in memory usage and a 2.3 times increase in throughput.