Conference paper Open Access

Transformers without Tears: Improving the Normalization of Self-Attention

Nguyen, Toan Q.; Salazar, Julian

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 l2 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 ob- serve 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.

Files (345.9 kB)
Name Size
IWSLT2019_paper_26.pdf
md5:5a6c18ef21719ddacdab79deec9a4b39
345.9 kB Download
0
0
views
downloads
All versions This version
Views 00
Downloads 00
Data volume 0 Bytes0 Bytes
Unique views 00
Unique downloads 00

Share

Cite as