analiza aiqbench
This arXiv paper examines how language models' internal representations affect next-token prediction. It finds that changes along the smallest singular directions of the unembedding matrix barely move the logits but still change predictions and loss more than random perturbations of equal size. For researchers working on interpretability, editing, or compression, this suggests that distance-based metrics can understate which internal directions actually matter. The findings are from open-weight models and remain to be tested more broadly.
The Smallest Singular Directions of the Unembedding Matter Disproportionately for Next-Token Prediction
arXiv:2605.17084v2 Announce Type: replace Abstract: Interpreting, editing and compressing language models often involve assessing which parts of their internal representations matter for prediction. Many of these assessments rely on distance, such as reconstruction error or the length of an intervention. For next-token prediction, however, the logit differences that matter most are those among the few most plausible tokens. These differences can be small. We therefore test whether changes to the final hidden state that move the logits less also change the prediction less. Specifically, changes along the smallest singular directions of the unembedding, which maps this state to the logits, move the logits less than random changes of the same length. In open-weight language models across several families, their prediction change and their change in the next-token loss are nevertheless on average several times those of random changes. Across all singular directions, the prediction change is out of proportion to the logit change only at the smallest. These directions carry a large share of the difference between the model's leading candidate tokens. Consequently, distance-based methods undervalue these directions, and rankings by singular value place them last, although they matter disproportionately for prediction. More broadly, our study highlights that comparing internal representations by distance can miss differences that change what the model predicts.
Abstract:Interpreting, editing and compressing language models often involve assessing which parts of their internal representations matter for prediction. Many of these assessments rely on distance, such as reconstruction error or the length of an intervention. For next-token prediction, however, the logit differences that matter most are those among the few most plausible tokens. These differences can be small. We therefore test whether changes to the final hidden state that move the logits less also change the prediction less. Specifically, changes along the smallest singular directions of the unembedding, which maps this state to the logits, move the logits less than random changes of the same length. In open-weight language models across several families, their prediction change and their change in the next-token loss are nevertheless on average several times those of random changes. Across all singular directions, the prediction change is out of proportion to the logit change only at the smallest. These directions carry a large share of the difference between the model's leading candidate tokens. Consequently, distance-based methods undervalue these directions, and rankings by singular value place them last, although they matter disproportionately for prediction. More broadly, our study highlights that comparing internal representations by distance can miss differences that change what the model predicts.
Machine Learning (cs.LG); Computation and Language (cs.CL)
arXiv:2605.17084 [cs.LG]
(or arXiv:2605.17084v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2605.17084 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 303 cuvinte
