SVDq: 1.25-bit and 410x Key Cache Compression for LLM Attention

TL;DR

SVDq is a novel SVD-based mixed precision quantization method that significantly compresses key caches in LLMs, achieving up to 410x compression with minimal performance loss.

Contribution

It introduces a new SVD-based quantization approach for key cache compression in LLMs, enabling ultra-low-bit precision with theoretical error bounds and high compression ratios.

Findings

Achieves 1.25-bit key cache precision with minimal loss.

Reaches up to 410x compression ratio when combined with sparsity.

Nearly lossless performance on LongBench datasets.

Abstract

For the efficient inference of Large Language Models (LLMs), the effective compression of key-value (KV) cache is essential. Three main types of KV cache compression techniques, namely sparsity, channel compression, and quantization, have been identified. This study presents SVDq, a Singular Value Decomposition (SVD) - based mixed precision quantization method for K cache. Initially, K cache is transformed into latent channels using SVD basis representations. Since the values in latent channels decay rapidly and become negligible after only a few latent channels, our method then incorporates importance-aware quantization and compression for latent channels. This enables the effective allocation of higher precision to more significant channels. Theoretically, we prove that SVDq results in quantization errors (x0.1 or even lower) that are much lower than those of per-channel key quantization in the original space. Our findings based on RULER and LongBench benchmarks demonstrate that SVDq can achieve an equivalent key cache precision as low as 1.25-bit. When combined with key sparsity, it can reach a key compression ratio of up to 410x for attention computation, all while maintaining comparable model performance. Notably, our method is nearly lossless for LongBench datasets. This indicates that SVDq enables high-precision low-bit quantization, providing a more efficient solution for KV cache compression in LLMs.