Ouroboros: On Accelerating Training of Transformer-Based Language Models

Abstract

Language models are essential for natural language processing (NLP) tasks, such as machine translation and text summarization. Remarkable performance has been demonstrated recently across many NLP domains via a Transformer-based language model with over a billion parameters, verifying the benefits of model size. Model parallelism is required if a model is too large to fit in a single computing device. Current methods for model parallelism either suffer from backward locking in backpropagation or are not applicable to language models. We propose the first model-parallel algorithm that speeds the training of Transformer-based language models. We also prove that our proposed algorithm is guaranteed to converge to critical points for non-convex problems. Extensive experiments on Transformer and Transformer-XL language models demonstrate that the proposed algorithm obtains a much faster speedup beyond data parallelism, with comparable or better accuracy. Code to reproduce experiments is to be found at \url{https://github.com/LaraQianYang/Ouroboros}.

arXiv commentTo appear in the proceedings of Neural Information Processing Systems Conference (2019)

References in corpus (10)

• Sequence to Sequence Learning with Neural Networks
• Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour
• SGDR: Stochastic Gradient Descent with Warm Restarts
• Training Deep Nets with Sublinear Memory Cost
• Pointer Sentinel Mixture Models
• Tying Word Vectors and Word Classifiers: A Loss Framework for Language Modeling
• A Deep Generative Framework for Paraphrase Generation
• Decoupled Parallel Backpropagation with Convergence Guarantee
• Training Neural Networks Using Features Replay
• Neural Machine Translation with Pivot Languages