DeBERTa: Decoding-enhanced BERT with Disentangled Attention
#deberta​ #bert​ #huggingface​ DeBERTa by Microsoft is the next iteration of BERT-style Self-Attention Transformer models, surpassing RoBERTa in State-of-the-art in multiple NLP tasks. DeBERTa brings two key improvements: First, they treat content and position information separately in a new form of disentangled attention mechanism. Second, they resort to relative positional encodings throughout the base of the transformer, and provide absolute positional encodings only at the very end. The resulting model is both more accurate on downstream tasks and needs less pretraining steps to reach good accuracy. Models are also available in Huggingface and on Github. 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 understanding (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, out performing the human baseline by a decent margin (90.3 versus 89.8). Authors: Pengcheng He, Xiaodong Liu, Jianfeng Gao, Weizhu Chen (Microsoft Dynamics 365 AI, Microsoft Research)

0:00​ - Intro & Overview 2:15​ - Position Encodings in Transformer's Attention Mechanism 9:55​ - Disentangling Content & Position Information in Attention 21:35​ - Disentangled Query & Key construction in the Attention Formula 25:50​ - Efficient Relative Position Encodings 28:40​ - Enhanced Mask Decoder using Absolute Position Encodings 35:30​ - My Criticism of EMD 38:05​ - Experimental Results 40:30​ - Scaling up to 1.5 Billion Parameters 44:20​ - Conclusion & Comments