LLM-Microscope: Uncovering the Hidden Role of Punctuation in Context Memory of Transformers
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2025
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866929723720859648 |
|---|---|
| author | Razzhigaev, Anton Mikhalchuk, Matvey Rahmatullaev, Temurbek Goncharova, Elizaveta Druzhinina, Polina Oseledets, Ivan Kuznetsov, Andrey |
| author_facet | Razzhigaev, Anton Mikhalchuk, Matvey Rahmatullaev, Temurbek Goncharova, Elizaveta Druzhinina, Polina Oseledets, Ivan Kuznetsov, Andrey |
| contents | We introduce methods to quantify how Large Language Models (LLMs) encode and store contextual information, revealing that tokens often seen as minor (e.g., determiners, punctuation) carry surprisingly high context. Notably, removing these tokens -- especially stopwords, articles, and commas -- consistently degrades performance on MMLU and BABILong-4k, even if removing only irrelevant tokens. Our analysis also shows a strong correlation between contextualization and linearity, where linearity measures how closely the transformation from one layer's embeddings to the next can be approximated by a single linear mapping. These findings underscore the hidden importance of filler tokens in maintaining context. For further exploration, we present LLM-Microscope, an open-source toolkit that assesses token-level nonlinearity, evaluates contextual memory, visualizes intermediate layer contributions (via an adapted Logit Lens), and measures the intrinsic dimensionality of representations. This toolkit illuminates how seemingly trivial tokens can be critical for long-range understanding. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_15007 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | LLM-Microscope: Uncovering the Hidden Role of Punctuation in Context Memory of Transformers Razzhigaev, Anton Mikhalchuk, Matvey Rahmatullaev, Temurbek Goncharova, Elizaveta Druzhinina, Polina Oseledets, Ivan Kuznetsov, Andrey Computation and Language Artificial Intelligence We introduce methods to quantify how Large Language Models (LLMs) encode and store contextual information, revealing that tokens often seen as minor (e.g., determiners, punctuation) carry surprisingly high context. Notably, removing these tokens -- especially stopwords, articles, and commas -- consistently degrades performance on MMLU and BABILong-4k, even if removing only irrelevant tokens. Our analysis also shows a strong correlation between contextualization and linearity, where linearity measures how closely the transformation from one layer's embeddings to the next can be approximated by a single linear mapping. These findings underscore the hidden importance of filler tokens in maintaining context. For further exploration, we present LLM-Microscope, an open-source toolkit that assesses token-level nonlinearity, evaluates contextual memory, visualizes intermediate layer contributions (via an adapted Logit Lens), and measures the intrinsic dimensionality of representations. This toolkit illuminates how seemingly trivial tokens can be critical for long-range understanding. |
| title | LLM-Microscope: Uncovering the Hidden Role of Punctuation in Context Memory of Transformers |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2502.15007 |