RealFormer: Transformer Likes Residual Attention

Code

• github.com/google-research/google-research
  OfficialIn papertf★ 37,519
• github.com/cloneofsimo/RealFormer-pytorch
  pytorch★ 101
• github.com/jaketae/realformer
  pytorch★ 11
• github.com/JunnYu/x-transformers-paddle
  jax★ 10
• github.com/aivolcano/BERT_MRC_CLS
  pytorch★ 5

Abstract

Transformer is the backbone of modern NLP models. In this paper, we propose RealFormer, a simple and generic technique to create Residual Attention Layer Transformer networks that significantly outperform the canonical Transformer and its variants (BERT, ETC, etc.) on a wide spectrum of tasks including Masked Language Modeling, GLUE, SQuAD, Neural Machine Translation, WikiHop, HotpotQA, Natural Questions, and OpenKP. We also observe empirically that RealFormer stabilizes training and leads to models with sparser attention. Source code and pre-trained checkpoints for RealFormer can be found at https://github.com/google-research/google-research/tree/master/realformer.

Tasks

Benchmark Results

Reproductions