Answer Fast: Accelerating BERT on the Tensor Streaming Processor

TL;DR

This paper presents a method to accelerate BERT inference on a tensor streaming processor by fusing nonlinear components with matrix multiplications, achieving a 6x speedup and deterministic low latency.

Contribution

It introduces a novel fusion technique for nonlinear and matrix components to optimize BERT inference on specialized hardware.

Findings

Deterministic tail latency of 130 μs for batch-1 BERT-base inference

6x faster inference compared to state-of-the-art methods

Efficient utilization of on-chip matrix multiplication units

Abstract

Transformers have become a predominant machine learning workload, they are not only the de-facto standard for natural language processing tasks, but they are also being deployed in other domains such as vision and speech recognition. Many of the transformer-based applications are real-time systems such as machine translation and web search. These real time systems often come with strict end-to-end inference latency requirements. Unfortunately, while the majority of the transformer computation comes from matrix multiplications, transformers also include several non-linear components that tend to become the bottleneck during an inference. In this work, we accelerate the inference of BERT models on the tensor streaming processor. By carefully fusing all the nonlinear components with the matrix multiplication components, we are able to efficiently utilize the on-chip matrix multiplication units resulting in a deterministic tail latency of 130 $\mu$s for a batch-1 inference through BERT-base, which is 6X faster than the current state-of-the-art.