DeLighT: Deep and Light-weight Transformer

We introduce a deep and light-weight transformer, DeLighT, that delivers similar or better performance than standard transformer-based models with significantly fewer parameters. DeLighT more efficiently allocates parameters both (1) within each Transformer block using the DeLighT transformation, a deep and light-weight transformation, and (2) across blocks using block-wise scaling, which allows for shallower and narrower DeLighT blocks near the input and wider and deeper DeLighT blocks near the output. Overall, DeLighT networks are 2.5 to 4 times deeper than standard transformer models and yet have fewer parameters and operations. Experiments on benchmark machine translation and language modeling tasks show that DeLighT matches or improves the performance of baseline Transformers with 2 to 3 times fewer parameters on average. Our source code is available at: \url{https://github.com/sacmehta/delight}

PDF Abstract ICLR 2021 PDF ICLR 2021 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Benchmark
Machine Translation IWSLT2014 German-English DeLighT BLEU score 35.3 # 22
Language Modelling WikiText-103 DeLighT Test perplexity 24.14 # 55
Number of params 99M # 42
Machine Translation WMT2016 English-French DeLighT BLEU score 40.5 # 1
Machine Translation WMT2016 English-German DeLighT BLEU score 28.0 # 4
Machine Translation WMT2016 English-Romanian DeLighT BLEU score 34.7 # 1

Methods