Совместная разработка Huawei предлагает новый взгляд на сжатие ИИ-моделей через прогрессивное низкоранговое разложение. Технология позволяет значительно облегчить архитектуру GPT-2, сохраняя высокую скорость генерации текста и точность работы трансформерных слоев.
Low rank decomposition decomposes each fully-connected layer of the transformer modules into two smaller layers using Singular Value Decomposition. The state-of-the-art techniques usually apply LRD in a single-shot, where all of the layers are decomposed simultaneously. In this paper, we propose and compare different strategies for applying low rank decomposition to compress pre-trained transformer based models. These strategies include: layer-by-layer and progressive decomposition. We observe that progressive low rank decomposition, in which the rank is decreased incrementally results in a higher accuracy after decomposition comparing to single-shot and layer-by-layer low rank decomposition. Furthermore, in contrast with many of state-of-the-art compression methods where intensive pre-training of the compressed model is necessary, we show that progressive LRD can provide promising performance by compressing the model in the fine-tuning stage.