LightSeq2: Accelerated Training for Transformer-based Models on GPUs

TL;DR

LightSeq2 is a GPU-optimized system that significantly accelerates training for various Transformer models, reducing training time and resource consumption across multiple architectures and benchmarks.

Contribution

The paper introduces LightSeq2, a novel GPU acceleration system specifically designed for diverse Transformer models, improving training speed and efficiency.

Findings

Achieves 1.4-3.5x faster training than previous systems.

Gains 308% speedup on WMT14 English-German translation benchmark.

Supports multiple Transformer architectures including BERT, GPT, and Vision Transformer.

Abstract

Transformer-based neural models are used in many AI applications. Training these models is expensive, as it takes huge GPU resources and long duration. It is challenging because typical data like sentences have variable lengths, and Transformer's computation patterns are more complex than convolutional neural networks. Existing systems either only focus on model inference or optimization for only BERT-like encoder models. In this paper, we present LightSeq2, a system to accelerate training for a general family of Transformer models on GPUs. We propose a series of GPU optimization techniques tailored to the specific computation flow and memory access patterns of Transformer models. LightSeq2 supports many model architectures, including BERT (encoder-only), GPT (decoder-only), Transformer (encoder-decoder), and vision Transformer. Our experiments for a variety of models and benchmarks show that LightSeq2 is consistently faster (1.4-3.5x) than previous systems on different GPUs. In particular, it gains 308% training speedup compared with existing systems on a large public machine translation benchmark (WMT14 English-German).