Transformer Feed-Forward Layers Build Predictions by Promoting Concepts in the Vocabulary Space
Open accessThe DOI matches its Crossref record; 17 of the 17 references it cites were found in a registry or an index.
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 reverseengineering 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. 1
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019) · Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) · doi:10.18653/v1/n19-1423 · Cited by 54,325
Language Models are Few-Shot Learners
Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al. (2020) · arXiv · doi:10.48550/arxiv.2005.14165 · Cited by 23,372
RoBERTa: A Robustly Optimized BERT Pretraining Approach
Liu, Y., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D., Levy, O., Lewis, M., Zettlemoyer, L., & Stoyanov, V. (2019) · arXiv · doi:10.48550/arxiv.1907.11692 · Cited by 16,515
Ridge regression: Biased estimation for nonorthogonal problems
Hoerl, A. E., & Kennard, R. W. (1970) · Technometrics · doi:10.1080/00401706.1970.10488634 · Cited by 9,648
BERT Rediscovers the Classical NLP Pipeline
Tenney, I., Das, D., & Pavlick, E. (2019) · Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/p19-1452 · Cited by 1,307
How Can We Know What Language Models Know?
Jiang, Z., Xu, F. F., Araki, J., & Neubig, G. (2020) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00324 · Cited by 1,002
RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models
Gehman, S., Gururangan, S., Sap, M., Choi, Y., & Smith, N. A. (2020) · Findings of the Association for Computational Linguistics: EMNLP 2020 · doi:10.18653/v1/2020.findings-emnlp.301 · Cited by 693
Transformer Feed-Forward Layers Are Key-Value Memories
Geva, M., Schuster, R., Berant, J., & Levy, O. (2020) · arXiv · doi:10.48550/arxiv.2012.14913 · Cited by 353
Self-citationTransformer Feed-Forward Layers Are Key-Value Memories
Geva, M., Schuster, R., Berant, J., & Levy, O. (2021) · Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/2021.emnlp-main.446 · Cited by 353
Self-citation
Cited by
Explainability for Large Language Models: A Survey
Zhao, H., Chen, H., Yang, F., Liu, N., Deng, H., Cai, H., Wang, S., Yin, D., & Du, M. (2024) · ACM Transactions on Intelligent Systems and Technology · doi:10.1145/3639372 · Cited by 681
Open Problems in Mechanistic Interpretability
Sharkey, L., Chughtai, B., Batson, J., Lindsey, J., Wu, J., Bushnaq, L., Goldowsky-Dill, N., Heimersheim, S., Ortega, A., Bloom, J., et al. (2025) · arXiv · doi:10.48550/arxiv.2501.16496 · Cited by 291
Self-citationThe Hydra Effect: Emergent Self-repair in Language Model Computations
McGrath, T., Rahtz, M., Kramar, J., Mikulik, V., & Legg, S. (2023) · arXiv · doi:10.48550/arxiv.2307.15771 · Cited by 140
Toward Transparent AI: A Survey on Interpreting the Inner Structures of Deep Neural Networks
Räuker, T., Ho, A., Casper, S., & Hadfield-Menell, D. (2023) · 2023 IEEE Conference on Secure and Trustworthy Machine Learning (SaTML) · doi:10.1109/satml54575.2023.00039 · Cited by 122
(2023) · doi:10.18653/v1/2023.emnlp-main.632 · Cited by 115
Metadata
- Authors
- Geva, M., Caciularu, A., Wang, K., & Goldberg, Y.
- Venue
- Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing
- Published
- 2022
- Pages
- 30-45
- OpenAlex
- W4385572928
- Cited by
- 117
- References
- 17
- FWCI
- 12.88
- Type
- conference paper
- Language
- en
- Open access
- hybrid
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Transformer Feed-Forward Layers Build Predictions by Promoting Concepts in the Vocabulary Space. https://app.cochat.ai/papers/10.18653/v1/2022.emnlp-main.3
- The paper
- Geva, M., Caciularu, A., Wang, K., & Goldberg, Y. (2022). Transformer Feed-Forward Layers Build Predictions by Promoting Concepts in the Vocabulary Space. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. https://doi.org/10.18653/v1/2022.emnlp-main.3
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.