English
Related papers

Related papers: Token Homogenization under Positional Bias

200 papers

Standard evaluations of Large language models (LLMs) focus on task performance, offering limited insight into whether correct behavior reflects appropriate underlying mechanisms and risking confirmation bias. We introduce a simple,…

Computation and Language · Computer Science 2026-04-01 Zoë Prins , Samuele Punzo , Frank Wildenburg , Giovanni Cinà , Sandro Pezzelle

Mechanisms for encoding positional information are central for transformer-based language models. In this paper, we analyze the position embeddings of existing language models, finding strong evidence of translation invariance, both for the…

Computation and Language · Computer Science 2021-06-04 Ulme Wennberg , Gustav Eje Henter

Homogeneity bias in Large Language Models (LLMs) refers to their tendency to homogenize the representations of some groups compared to others. Previous studies documenting this bias have predominantly used encoder models, which may have…

Computation and Language · Computer Science 2024-12-13 Messi H. J. Lee , Calvin K. Lai

Token representation strategies within large-scale neural architectures often rely on contextually refined embeddings, yet conventional approaches seldom encode structured relationships explicitly within token interactions. Self-attention…

Computation and Language · Computer Science 2025-03-27 James Blades , Frederick Somerfield , William Langley , Susan Everingham , Maurice Witherington

Recent research suggests that the feed-forward module within Transformers can be viewed as a collection of key-value memories, where the keys learn to capture specific patterns from the input based on the training examples. The values then…

Computation and Language · Computer Science 2023-10-25 Sunit Bhattacharya , Ondrej Bojar

Large Language Models (LLMs) have made significant strides in natural language processing, and a precise understanding of the internal mechanisms driving their success is essential. In this work, we analyze the trajectories of token…

Machine Learning · Computer Science 2025-03-06 Murdock Aubry , Haoming Meng , Anton Sugolov , Vardan Papyan

Modern tokenizers employ deterministic algorithms to map text into a single "canonical" token sequence, yet the same string can be encoded as many non-canonical tokenizations using the tokenizer vocabulary. In this work, we investigate the…

Computation and Language · Computer Science 2026-02-04 Brian Siyuan Zheng , Alisa Liu , Orevaoghene Ahia , Jonathan Hayase , Yejin Choi , Noah A. Smith

Adapting language models to new data distributions by simple finetuning is challenging. This is due to the rigidity of their subword tokenizers, which typically remain unchanged during adaptation. This inflexibility often leads to…

Computation and Language · Computer Science 2026-05-14 Abraham Toluwase Owodunni , Orevaoghene Ahia , Sachin Kumar

Tokenization is a fundamental component of large language models (LLMs), yet its influence on model scaling and performance is not fully explored. In this paper, we introduce Over-Tokenized Transformers, a novel framework that decouples…

Computation and Language · Computer Science 2025-05-26 Hongzhi Huang , Defa Zhu , Banggu Wu , Yutao Zeng , Ya Wang , Qiyang Min , Xun Zhou

Tokenization is a foundational step in the text process of Large Language Models (LLMs). Texts must be first tokenized into token IDs, which are then input to LLMs. Inefficient tokenization results in long token-ID sequences and will slow…

Computation and Language · Computer Science 2026-05-14 Chong Li , Yingzhuo Deng , Wen Yang , Jiajun Zhang , Chengqing Zong

Transformers underpin modern large language models (LLMs) and are commonly assumed to be behaviorally unstructured at random initialization, with all meaningful preferences emerging only through large-scale training. We challenge this…

Machine Learning · Statistics 2026-02-06 Siquan Li , Yao Tong , Haonan Wang , Tianyang Hu

Padding tokens are widely used in large language models (LLMs) to equalize sequence lengths during batched inference. While they should be fully masked, implementation errors can cause them to influence computation, and the extent of this…

Computation and Language · Computer Science 2025-10-07 Rom Himelstein , Amit LeVi , Yonatan Belinkov , Avi Mendelson

We explore the role of torsion in hybrid deep learning models that incorporate topological data analysis, focusing on autoencoders. While most TDA tools use field coefficients, this conceals torsional features present in integer homology.…

Algebraic Topology · Mathematics 2025-09-22 Maria Walch

Tokenization and transfer learning are two critical components in building state of the art time series foundation models for forecasting. In this work, we systematically study the effect of tokenizer design, specifically scaling and…

Machine Learning · Computer Science 2025-11-18 Alexis Roger , Gwen Legate , Kashif Rasul , Yuriy Nevmyvaka , Irina Rish

Causal self-attention provides positional information to Transformer decoders. Prior work has shown that stacks of causal self-attention layers alone induce a positional bias in attention scores toward earlier tokens. However, this differs…

Computation and Language · Computer Science 2026-04-21 Junu Kim , Xiao Liu , Zhenghao Lin , Lei Ji , Yeyun Gong , Edward Choi

In the paper we show that there is an analogy between the operations occurring in a layer of a transformer (projections and layer normalizations, disregarding the feedforward neural network) and a step in the power method. Coherently with…

Machine Learning · Computer Science 2026-05-26 Chenglong Li , Claudio Altafini

Large Vision Language Models (LVLMs) have demonstrated remarkable capabilities in understanding and describing visual content, achieving state-of-the-art performance across various vision-language tasks. However, these models often generate…

Computer Vision and Pattern Recognition · Computer Science 2025-03-27 Kazi Hasan Ibn Arif , Sajib Acharjee Dip , Khizar Hussain , Lang Zhang , Chris Thomas

Large Language Models (LLMs) are typically shipped with tokenizers that deterministically encode text into so-called canonical token sequences, to which the LLMs assign probability values. One common assumption is that the probability of a…

Computation and Language · Computer Science 2025-06-09 Renato Lui Geh , Honghua Zhang , Kareem Ahmed , Benjie Wang , Guy Van den Broeck

We take a physics-based approach to studying how the internal hidden states of large language models change from token to token during inference. Across 20 open-source transformer models (135M-3B parameters), we find that a quantity…

Computation and Language · Computer Science 2025-08-13 Lorenzo Tomaz , Judd Rosenblatt , Thomas Berry Jones , Diogo Schwerz de Lucena

Tokenization is widely used in large language models because it significantly improves performance. However, tokenization imposes several disadvantages, such as performance biases, increased adversarial vulnerability, decreased…

Computation and Language · Computer Science 2024-10-08 Kevin Slagle