QUICK: Quantization-aware Interleaving and Conflict-free Kernel for efficient LLM inference

arXiv:2402.10076v1 Announce Type: new
Abstract: We introduce QUICK, a group of novel optimized CUDA kernels for the efficient inference of quantized Large Language Models (LLMs). QUICK addresses the shared memory bank-conflict problem of state-of-the-art mixed precision matrix multiplication kernels. Our method interleaves the quantized weight matrices of LLMs offline to skip the shared memory write-back after the dequantization. We demonstrate up to 1.91x speedup over existing kernels of AutoAWQ on larger batches and up to 1.94x throughput gain on representative …

abstract art arxiv bank conflict cs.ai cs.lg cuda free inference interleaving kernel language language models large language large language models llm llms matrix matrix multiplication memory mixed novel offline precision quantization state type