Communication Compression for Tensor Parallel LLM Inference

Jan Hansen-Palmus, Michael Truong Le, Oliver Hausdörfer, Alok Verma

Keywords: Compression, Tensor Parallel, LLM, Inference, Quantization

TL;DR: Paper proposes to improve LLM inference latency by compressing inter-accelerator communication through block-wise quantization.

Abstract: Large Language Models (LLMs) have pushed the frontier of artificial intelligence but are comprised of hundreds of billions of parameters and operations. For faster inference latency, LLMs are deployed on multiple hardware accelerators through various Model Parallelism strategies. Our paper looks into the details on one such strategy - Tensor Parallelism - and proposes to reduce latency by compressing inter-accelerator communication. We leverage fine grained quantization techniques to compress selected activations between 3.5 - 4.5x. Our proposed method leads up to 2x reduction of time-to-first-token (TTFT) with negligible model performance degradation.

Submission Number: 107