DeLighT: Deep and Light-weight Transformer

TL;DR

DeLighT is a novel deep, light-weight transformer architecture that achieves comparable or superior performance to standard models while significantly reducing parameters and computational requirements, through efficient parameter allocation within and across blocks.

Contribution

This paper introduces DeLighT, a transformer design that is deeper and more parameter-efficient, with innovative intra- and inter-block parameter scaling strategies.

Findings

DeLighT matches or outperforms baseline transformers on benchmarks.

DeLighT uses 2-3 times fewer parameters.

DeLighT networks are 2.5 to 4 times deeper than standard transformers.

Abstract

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}