Subformer: Exploring Weight Sharing for Parameter Efficiency in Generative Transformers
This work addresses the computational expense and high parameter budget of Transformers for researchers and practitioners working with generative models, offering an incremental improvement in parameter efficiency.
This paper explores parameter-sharing methods in generative Transformers to address computational expense and high parameter budgets. The authors developed the Subformer, which combines sandwich-style parameter sharing and self-attentive embedding factorization (SAFE), demonstrating that it can outperform the Transformer with significantly fewer parameters on tasks like machine translation, abstractive summarization, and language modeling.
Transformers have shown improved performance when compared to previous architectures for sequence processing such as RNNs. Despite their sizeable performance gains, as recently suggested, the model is computationally expensive to train and with a high parameter budget. In light of this, we explore parameter-sharing methods in Transformers with a specific focus on generative models. We perform an analysis of different parameter sharing/reduction methods and develop the Subformer. Our model combines sandwich-style parameter sharing, which overcomes naive cross-layer parameter sharing in generative models, and self-attentive embedding factorization (SAFE). Experiments on machine translation, abstractive summarization and language modeling show that the Subformer can outperform the Transformer even when using significantly fewer parameters.