Sparse Upcycling: Training Mixture-of-Experts from Dense Checkpoints
This reduces training costs for large-scale sparse models, benefiting researchers and practitioners in NLP and vision, though it is incremental as it builds on existing dense and sparse model techniques.
The paper tackles the high cost of training sparse models by proposing sparse upcycling, which initializes Mixture-of-Experts models from dense checkpoints, resulting in upcycled models outperforming dense counterparts on SuperGLUE and ImageNet using only about 50% of the initial training cost.
Training large, deep neural networks to convergence can be prohibitively expensive. As a result, often only a small selection of popular, dense models are reused across different contexts and tasks. Increasingly, sparsely activated models, which seek to decouple model size from computation costs, are becoming an attractive alternative to dense models. Although more efficient in terms of quality and computation cost, sparse models remain data-hungry and costly to train from scratch in the large scale regime. In this work, we propose sparse upcycling -- a simple way to reuse sunk training costs by initializing a sparsely activated Mixture-of-Experts model from a dense checkpoint. We show that sparsely upcycled T5 Base, Large, and XL language models and Vision Transformer Base and Large models, respectively, significantly outperform their dense counterparts on SuperGLUE and ImageNet, using only ~50% of the initial dense pretraining sunk cost. The upcycled models also outperform sparse models trained from scratch on 100% of the initial dense pretraining computation budget.