DEBERTA: DECODING-ENHANCED BERT WITH DISENTANGLED ATTENTION

Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen

Keywords: Transformer, Attention, Natural Language Processing, Language Model Pre-training, Position Encoding

Abstract:

Recent progress in pre-trained neural language models has significantly improved the performance of many natural language processing (NLP) tasks. In this paper we propose a new model architecture DeBERTa (Decoding-enhanced BERT with disentangled attention) that improves the BERT and RoBERTa models using two novel techniques. The first is the disentangled attention mechanism, where each word is represented using two vectors that encode its content and position, respectively, and the attention weights among words are computed using disentangled matrices on their contents and relative positions, respectively. Second, an enhanced mask decoder is used to incorporate absolute positions in the decoding layer to predict the masked tokens in model pre-training. In addition, a new virtual adversarial training method is used for fine-tuning to improve models’
generalization. We show that these techniques significantly improve the efficiency of model pre-training and the performance of both natural language understand(NLU) and natural langauge generation (NLG) downstream tasks. Compared to RoBERTa-Large, a DeBERTa model trained on half of the training data performs consistently better on a wide range of NLP tasks, achieving improvements on MNLI by +0.9% (90.2% vs. 91.1%), on SQuAD v2.0 by +2.3% (88.4% vs. 90.7%) and RACE by +3.6% (83.2% vs. 86.8%). Notably, we scale up DeBERTa by training a larger version that consists of 48 Transform layers with 1.5 billion parameters. The significant performance boost makes the single DeBERTa model surpass the human performance on the SuperGLUE benchmark (Wang et al., 2019a) for the first time in terms of macro-average score (89.9 versus 89.8), and the ensemble DeBERTa model sits atop the SuperGLUE leaderboard as of January 6, 2021, outperforming the human baseline by a decent margin (90.3 versus
89.8). The pre-trained DeBERTa models and the source code were released at: https://github.com/microsoft/DeBERTa.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

One-sentence Summary: A new model architecture DeBERTa is proposed that improves the BERT and RoBERTa models using disentangled attention and enhanced mask decoder.

Code: [ microsoft/DeBERTa](https://github.com/microsoft/DeBERTa) + [ 8 community implementations](https://paperswithcode.com/paper/?openreview=XPZIaotutsD)

Data: [BookCorpus](https://paperswithcode.com/dataset/bookcorpus), [BoolQ](https://paperswithcode.com/dataset/boolq), [COPA](https://paperswithcode.com/dataset/copa), [CoLA](https://paperswithcode.com/dataset/cola), [CommitmentBank](https://paperswithcode.com/dataset/commitmentbank), [GLUE](https://paperswithcode.com/dataset/glue), [HellaSwag](https://paperswithcode.com/dataset/hellaswag), [MRPC](https://paperswithcode.com/dataset/mrpc), [MultiNLI](https://paperswithcode.com/dataset/multinli), [MultiRC](https://paperswithcode.com/dataset/multirc), [QNLI](https://paperswithcode.com/dataset/qnli), [Quora Question Pairs](https://paperswithcode.com/dataset/quora-question-pairs), [RACE](https://paperswithcode.com/dataset/race), [RTE](https://paperswithcode.com/dataset/rte), [ReCoRD](https://paperswithcode.com/dataset/record), [SQuAD](https://paperswithcode.com/dataset/squad), [SST](https://paperswithcode.com/dataset/sst), [SST-2](https://paperswithcode.com/dataset/sst-2), [STS Benchmark](https://paperswithcode.com/dataset/sts-benchmark), [SWAG](https://paperswithcode.com/dataset/swag), [SuperGLUE](https://paperswithcode.com/dataset/superglue), [WNLI](https://paperswithcode.com/dataset/wnli), [WSC](https://paperswithcode.com/dataset/wsc)

Community Implementations: [ 7 code implementations](https://www.catalyzex.com/paper/arxiv:2006.03654/code)