SparQ Attention: Bandwidth-Efficient LLM Inference

TL;DR

SparQ Attention enhances large language model inference efficiency by reducing data transfer bottlenecks through selective memory fetching, enabling faster processing without retraining or fine-tuning.

Contribution

It introduces a novel attention technique that improves inference throughput by optimizing memory bandwidth usage, applicable to existing LLMs without retraining.

Findings

Up to 8x reduction in attention data transfers

Minimal accuracy loss across multiple models and tasks

Applicable to off-the-shelf LLMs without modifications

Abstract

The computational difficulties of large language model (LLM) inference remain a significant obstacle to their widespread deployment. The need for many applications to support long input sequences and process them in large batches typically causes token-generation to be bottlenecked by data transfer. For this reason, we introduce SparQ Attention, a technique for increasing the inference throughput of LLMs by utilising memory bandwidth more efficiently within the attention layers, through selective fetching of the cached history. Our proposed technique can be applied directly to off-the-shelf LLMs during inference, without requiring any modification to the pre-training setup or additional fine-tuning. We show that SparQ Attention brings up to 8x savings in attention data transfers without substantial drops in accuracy, by evaluating Llama 2 and 3, Mistral, Gemma and Pythia models on a wide range of downstream tasks.