Transformers without Tears: Improving the Normalization of Self-Attention
This work addresses training instability and efficiency issues in Transformers for machine translation, particularly in low-resource settings, though it is incremental with mixed results in high-resource scenarios.
The paper tackled improving Transformer training stability and performance by evaluating three normalization-centric changes, resulting in an average +1.1 BLEU gain over state-of-the-art baselines on low-resource translation tasks and a new 32.8 BLEU score on IWSLT'15 English-Vietnamese.
We evaluate three simple, normalization-centric changes to improve Transformer training. First, we show that pre-norm residual connections (PreNorm) and smaller initializations enable warmup-free, validation-based training with large learning rates. Second, we propose $\ell_2$ normalization with a single scale parameter (ScaleNorm) for faster training and better performance. Finally, we reaffirm the effectiveness of normalizing word embeddings to a fixed length (FixNorm). On five low-resource translation pairs from TED Talks-based corpora, these changes always converge, giving an average +1.1 BLEU over state-of-the-art bilingual baselines and a new 32.8 BLEU on IWSLT'15 English-Vietnamese. We observe sharper performance curves, more consistent gradient norms, and a linear relationship between activation scaling and decoder depth. Surprisingly, in the high-resource setting (WMT'14 English-German), ScaleNorm and FixNorm remain competitive but PreNorm degrades performance.