Abstract
Transformer-based language models (LMs) are at the core of modern NLP, but their internal prediction construction process is opaque and largely not understood. In this work, we make a substantial step towards unveiling this underlying prediction process, by reverse-engineering the operation of the feed-forward network (FFN) layers, one of the building blocks of transformer models. We view the token representation as a changing distribution over the vocabulary, and the output from each FFN layer as an additive update to that distribution. Then, we analyze the FFN updates in the vocabulary space, showing that each update can be decomposed to sub-updates corresponding to single FFN parameter vectors, each promoting concepts that are often human-interpretable. We then leverage these findings for controlling LM predictions, where we reduce the toxicity of GPT2 by almost 50%, and for improving computation efficiency with a simple early exit rule, saving 20% of computation on average.
Original language | English |
---|---|
Pages | 30-45 |
Number of pages | 16 |
State | Published - 2022 |
Event | 2022 Conference on Empirical Methods in Natural Language Processing, EMNLP 2022 - Abu Dhabi, United Arab Emirates Duration: 7 Dec 2022 → 11 Dec 2022 |
Conference
Conference | 2022 Conference on Empirical Methods in Natural Language Processing, EMNLP 2022 |
---|---|
Country/Territory | United Arab Emirates |
City | Abu Dhabi |
Period | 7/12/22 → 11/12/22 |
Bibliographical note
Publisher Copyright:© 2022 Association for Computational Linguistics.