Outlier Dimensions that Disrupt Transformers are Driven by Frequency
Giovanni Puccetti, Anna Rogers, Aleksandr Drozd, Felice Dell’Orletta
Abstract
While Transformer-based language models are generally very robust to pruning, there is the recently discovered outlier phenomenon: disabling only 48 out of 110M parameters in BERT-base drops its performance by nearly 30% on MNLI. We replicate the original evidence for the outlier phenomenon and we link it to the geometry of the embedding space. We find that in both BERT and RoBERTa the magnitude of hidden state coefficients corresponding to outlier dimensions correlate with the frequencies of encoded tokens in pre-training data, and they also contribute to the “vertical” self-attention pattern enabling the model to focus on the special tokens. This explains the drop in performance from disabling the outliers, and it suggests that to decrease anisotopicity in future models we need pre-training schemas that would better take into account the skewed token distributions.- Anthology ID:
- 2022.findings-emnlp.93
- Volume:
- Findings of the Association for Computational Linguistics: EMNLP 2022
- Month:
- December
- Year:
- 2022
- Address:
- Abu Dhabi, United Arab Emirates
- Editors:
- Yoav Goldberg, Zornitsa Kozareva, Yue Zhang
- Venue:
- Findings
- SIG:
- Publisher:
- Association for Computational Linguistics
- Note:
- Pages:
- 1286–1304
- Language:
- URL:
- https://aclanthology.org/2022.findings-emnlp.93
- DOI:
- 10.18653/v1/2022.findings-emnlp.93
- Cite (ACL):
- Giovanni Puccetti, Anna Rogers, Aleksandr Drozd, and Felice Dell’Orletta. 2022. Outlier Dimensions that Disrupt Transformers are Driven by Frequency. In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 1286–1304, Abu Dhabi, United Arab Emirates. Association for Computational Linguistics.
- Cite (Informal):
- Outlier Dimensions that Disrupt Transformers are Driven by Frequency (Puccetti et al., Findings 2022)
- PDF:
- https://preview.aclanthology.org/ingest-acl-2023-videos/2022.findings-emnlp.93.pdf