On Layer Normalization in the Transformer Architecture
This addresses a practical bottleneck for researchers and practitioners in NLP by reducing training complexity and time, though it is incremental as it builds on existing Transformer modifications.
The paper tackles the problem of needing a learning rate warm-up stage for training Transformers, which slows optimization and increases hyper-parameter tuning, by showing that placing layer normalization inside residual blocks (Pre-LN) allows removal of the warm-up stage, achieving comparable results with significantly less training time and tuning across applications.
The Transformer is widely used in natural language processing tasks. To train a Transformer however, one usually needs a carefully designed learning rate warm-up stage, which is shown to be crucial to the final performance but will slow down the optimization and bring more hyper-parameter tunings. In this paper, we first study theoretically why the learning rate warm-up stage is essential and show that the location of layer normalization matters. Specifically, we prove with mean field theory that at initialization, for the original-designed Post-LN Transformer, which places the layer normalization between the residual blocks, the expected gradients of the parameters near the output layer are large. Therefore, using a large learning rate on those gradients makes the training unstable. The warm-up stage is practically helpful for avoiding this problem. On the other hand, our theory also shows that if the layer normalization is put inside the residual blocks (recently proposed as Pre-LN Transformer), the gradients are well-behaved at initialization. This motivates us to remove the warm-up stage for the training of Pre-LN Transformers. We show in our experiments that Pre-LN Transformers without the warm-up stage can reach comparable results with baselines while requiring significantly less training time and hyper-parameter tuning on a wide range of applications.