CodeGEMM: A Codebook-Centric Approach to Efficient GEMM in Quantized LLMs

TL;DR

CodeGEMM introduces a codebook-centric GEMM kernel that significantly accelerates quantized LLM inference by replacing dequantization with precomputed partial sums, enhancing speed and efficiency.

Contribution

It proposes a novel GEMM kernel that eliminates dequantization overhead using precomputed inner products, enabling faster and more efficient quantized LLM inference.

Findings

Achieves 1.83x speedup on 8B Llama-3 models in 2-bit quantization.

Achieves 8.93x speedup on 70B Llama-3 models in 2-bit quantization.

Reduces latency and cache pressure compared to existing methods.

Abstract

Weight-only quantization is widely used to mitigate the memory-bound nature of LLM inference. Codebook-based methods extend this trend by achieving strong accuracy in the extremely low-bit regime (e.g., 2-bit). However, current kernels rely on dequantization, which repeatedly fetches centroids and reconstructs weights, incurring substantial latency and cache pressure. We present CodeGEMM, a codebook-centric GEMM kernel that replaces dequantization with precomputed inner products between centroids and activations stored in a lightweight Psumbook. At inference, code indices directly gather these partial sums, eliminating per-element lookups and reducing the on-chip footprint. The kernel supports the systematic exploration of latency-memory-accuracy trade-offs under a unified implementation. On Llama-3 models, CodeGEMM delivers 1.83x (8B) and 8.93x (70B) speedups in the 2-bit configuration compared to state-of-the-art codebook-based quantization at comparable accuracy and further improves computing efficiency and memory subsystem utilization.