The Role of Sparsity for Length Generalization in Transformers

• URL: http://arxiv.org/abs/2502.16792v1
• Date: Mon, 24 Feb 2025 03:01:03 GMT
• Title: The Role of Sparsity for Length Generalization in Transformers
• Authors: Noah Golowich, Samy Jelassi, David Brandfonbrener, Sham M. Kakade, Eran Malach,
• Abstract summary: We propose a new theoretical framework to study length generalization for the next-token prediction task.<n>We show that length generalization occurs as long as each predicted token depends on a small (fixed) number of previous tokens.<n>We introduce Predictive Position Coupling, which trains the transformer to predict the position IDs used in a positional coupling approach.
• Score: 58.65997625433689
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Training large language models to predict beyond their training context lengths has drawn much attention in recent years, yet the principles driving such behavior of length generalization remain underexplored. We propose a new theoretical framework to study length generalization for the next-token prediction task, as performed by decoder-only transformers. Conceptually, we show that length generalization occurs as long as each predicted token depends on a small (fixed) number of previous tokens. We formalize such tasks via a notion we call $k$-sparse planted correlation distributions, and show that an idealized model of transformers which generalize attention heads successfully length-generalize on such tasks. As a bonus, our theoretical model justifies certain techniques to modify positional embeddings which have been introduced to improve length generalization, such as position coupling. We support our theoretical results with experiments on synthetic tasks and natural language, which confirm that a key factor driving length generalization is a ``sparse'' dependency structure of each token on the previous ones. Inspired by our theory, we introduce Predictive Position Coupling, which trains the transformer to predict the position IDs used in a positional coupling approach. Predictive Position Coupling thereby allows us to broaden the array of tasks to which position coupling can successfully be applied to achieve length generalization.

Related papers

• Beyond Progress Measures: Theoretical Insights into the Mechanism of Grokking [50.465604300990904]
  Grokking refers to the abrupt improvement in test accuracy after extended overfitting. In this work, we investigate the grokking mechanism underlying the Transformer in the task of prime number operations.
  arXiv  Detail & Related papers  (2025-04-04T04:42:38Z)
• Arithmetic Transformers Can Length-Generalize in Both Operand Length and Count [19.148785141454642]
  Transformers often struggle with length generalization, meaning they fail to generalize to sequences longer than those encountered during training. In this work, we achieve approximately 2-3x length generalization on both tasks, which is the first such achievement in arithmetic Transformers.
  arXiv  Detail & Related papers  (2024-10-21T08:49:51Z)
• A Formal Framework for Understanding Length Generalization in Transformers [14.15513446489798]
  We introduce a rigorous theoretical framework to analyze length generalization in causal transformers. We experimentally validate the theory as a predictor of success and failure of length generalization across a range of algorithmic and formal language tasks.
  arXiv  Detail & Related papers  (2024-10-03T01:52:01Z)
• Length Generalization of Causal Transformers without Position Encoding [59.802708262402824]
  Generalizing to longer sentences is important for recent Transformer-based language models. We study the length generalization property of Transformers without position encodings. We find that although NoPE can extend to sequences longer than the commonly used explicit position encodings, it still has a limited context length.
  arXiv  Detail & Related papers  (2024-04-18T14:38:32Z)
• Transformers Can Achieve Length Generalization But Not Robustly [76.06308648699357]
  We show that the success of length generalization is intricately linked to the data format and the type of position encoding. We show for the first time that standard Transformers can extrapolate to a sequence length that is 2.5x the input length.
  arXiv  Detail & Related papers  (2024-02-14T18:18:29Z)
• On Provable Length and Compositional Generalization [7.883808173871223]
  Out-of-distribution generalization capabilities of sequence-to-sequence models can be studied from the lens of two crucial forms of generalization.<n>We provide first provable guarantees on length and compositional generalization for common sequence-to-sequence models.
  arXiv  Detail & Related papers  (2024-02-07T14:16:28Z)
• What Algorithms can Transformers Learn? A Study in Length Generalization [23.970598914609916]
  We study the scope of Transformers' abilities in the specific setting of length generalization on algorithmic tasks. Specifically, we leverage RASP -- a programming language designed for the computational model of a Transformer. Our work provides a novel perspective on the mechanisms of compositional generalization and the algorithmic capabilities of Transformers.
  arXiv  Detail & Related papers  (2023-10-24T17:43:29Z)
• From Interpolation to Extrapolation: Complete Length Generalization for Arithmetic Transformers [7.011373967209572]
  We show that transformer models are able to generalize to long lengths with the help of targeted attention biasing. We demonstrate that using ABC, the transformer model can achieve unprecedented near-perfect length generalization on certain arithmetic tasks.
  arXiv  Detail & Related papers  (2023-10-18T14:10:47Z)
• Improving Length-Generalization in Transformers via Task Hinting [42.95479331339189]
  In particular, the performance of a transformer model trained on tasks up to a certain length drops sharply when applied to longer instances of the same problem. This work proposes an approach based on task hinting towards addressing length generalization.
  arXiv  Detail & Related papers  (2023-10-01T16:57:40Z)
• Disentangled Sequence to Sequence Learning for Compositional Generalization [62.954842223732435]
  We propose an extension to sequence-to-sequence models which allows us to learn disentangled representations by adaptively re-encoding the source input. Experimental results on semantic parsing and machine translation empirically show that our proposal yields more disentangled representations and better generalization.
  arXiv  Detail & Related papers  (2021-10-09T22:27:19Z)
• Predicting Deep Neural Network Generalization with Perturbation Response Curves [58.8755389068888]
  We propose a new framework for evaluating the generalization capabilities of trained networks. Specifically, we introduce two new measures for accurately predicting generalization gaps. We attain better predictive scores than the current state-of-the-art measures on a majority of tasks in the Predicting Generalization in Deep Learning (PGDL) NeurIPS 2020 competition.
  arXiv  Detail & Related papers  (2021-06-09T01:37:36Z)
• Measuring Generalization with Optimal Transport [111.29415509046886]
  We develop margin-based generalization bounds, where the margins are normalized with optimal transport costs. Our bounds robustly predict the generalization error, given training data and network parameters, on large scale datasets.
  arXiv  Detail & Related papers  (2021-06-07T03:04:59Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.