Outliers Dimensions that Disrupt Transformers Are Driven by Frequency
By Giovanni Puccetti, Anna Rogers, Aleksandr Drozd, Felice Dell'Orletta. 2022
65 citations recorded by the source corpus.
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 correlates with the frequency of encoded tokens in pre-training data, and it also contributes 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 anisotropicity in future models we need pre-training schemas that would better take into account the skewed token distributions.
Subfields: Electrical and Electronic Engineering
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Electrical and Electronic Engineering
Citations in the corpus
- The case for 4-bit precision: k-bit Inference Scaling Laws (2022, background): “…language models (Gao et al., 2019; Timkey & van Schijndel, 2021; Bondarenko et al., 2021; Wei et al., 2022; Luo et al., 2021; Kovaleva et al., 2021; Puccetti et al., 2022) can cause large quantization errors and severe performance degradation (Dettmers et al., 2022a; Zeng et al., 2022; Xiao et…”
- LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale (2022, neutral): “Similarly, Kovaleva et al. (2021) attribute the appearance of outliers in BERT model family to LayerNorm, and Puccetti et al. (2022) show empirically that outlier emergence is related to the frequency of tokens in the training distribution.”