English

FastSeq: Make Sequence Generation Faster

Computation and Language 2021-07-14 v2 Machine Learning

Abstract

Transformer-based models have made tremendous impacts in natural language generation. However the inference speed is a bottleneck due to large model size and intensive computing involved in auto-regressive decoding process. We develop FastSeq framework to accelerate sequence generation without accuracy loss. The proposed optimization techniques include an attention cache optimization, an efficient algorithm for detecting repeated n-grams, and an asynchronous generation pipeline with parallel I/O. These optimizations are general enough to be applicable to Transformer-based models (e.g., T5, GPT2, and UniLM). Our benchmark results on a set of widely used and diverse models demonstrate 4-9x inference speed gain. Additionally, FastSeq is easy to use with a simple one-line code change. The source code is available at https://github.com/microsoft/fastseq.

Keywords

Cite

@article{arxiv.2106.04718,
  title  = {FastSeq: Make Sequence Generation Faster},
  author = {Yu Yan and Fei Hu and Jiusheng Chen and Nikhil Bhendawade and Ting Ye and Yeyun Gong and Nan Duan and Desheng Cui and Bingyu Chi and Ruofei Zhang},
  journal= {arXiv preprint arXiv:2106.04718},
  year   = {2021}
}

Comments

ACL 2021 Demo Track