FFN SwiGLU
14 February 2020
FFN SwiGLU from Google, language model published 2020-02-14. Notable for: historical significance.
Abstract
Gated Linear Units (arXiv:1612.08083) consist of the component-wise product of two linear projections, one of which is first passed through a sigmoid function. Variations on GLU are possible, using different nonlinear (or even linear) functions in place of sigmoid. We test these variants in the feed-forward sublayers of the Transformer (arXiv:1706.03762) sequence-to-sequence model, and find that some of them yield quality improvements over the typically-used ReLU or GELU activations.
Sources: Epoch AI. This entry was generated from a dataset and has not been edited by hand.