English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Transformer-based language models (LMs) are at the core of modern NLP, buttheir internal prediction construction process is opaque and largely notunderstood. In this work, we make a substantial step towards unveiling thisunderlying prediction process, by reverse-engineering the operation of thefeed-forward network (FFN) layers, one of the building blocks of transformermodels. We view the token representation as a changing distribution over thevocabulary, and the output from each FFN layer as an additive update to thatdistribution. Then, we analyze the FFN updates in the vocabulary space, showingthat each update can be decomposed to sub-updates corresponding to single FFNparameter vectors, each promoting concepts that are often human-interpretable.We then leverage these findings for controlling LM predictions, where we reducethe toxicity of GPT2 by almost 50%, and for improving computation efficiencywith a simple early exit rule, saving 20% of computation on average.

Transformer Feed-Forward Layers Build Predictions by Promoting Concepts in the Vocabulary Space