Length Generalization Bounds for Transformers

• URL: http://arxiv.org/abs/2603.02238v1
• Date: Fri, 13 Feb 2026 14:49:27 GMT
• Title: Length Generalization Bounds for Transformers
• Abstract summary: Length generalization is a key property of a learning algorithm that enables it to make correct predictions on inputs of any length.<n>To provide such a guarantee, one needs to be able to compute a length generalization bound, beyond which the model is guaranteed to generalize.<n>This paper concerns the open problem of the computability of such generalization bounds for CRASP, a class of languages which is closely linked to transformers.
• Score: 9.186207021992535
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Length generalization is a key property of a learning algorithm that enables it to make correct predictions on inputs of any length, given finite training data. To provide such a guarantee, one needs to be able to compute a length generalization bound, beyond which the model is guaranteed to generalize. This paper concerns the open problem of the computability of such generalization bounds for CRASP, a class of languages which is closely linked to transformers. A positive partial result was recently shown by Chen et al. for CRASP with only one layer and, under some restrictions, also with two layers. We provide complete answers to the above open problem. Our main result is the non-existence of computable length generalization bounds for CRASP (already with two layers) and hence for transformers. To complement this, we provide a computable bound for the positive fragment of CRASP, which we show equivalent to fixed-precision transformers. For both positive CRASP and fixed-precision transformers, we show that the length complexity is exponential, and prove optimality of the bounds.

Related papers

• Revisiting Padded Transformer Expressivity: Which Architectural Choices Matter and Which Don't [78.36679389968772]
  We find that padded $textL$ constantprecision transformers are surprisingly robust to expressivity.<n>$logd-looped constant-precision transformers reach $textFO-uniform TCd$, and growing-precision ones reach $textFO-uniform TCd$.
  arXiv  Detail & Related papers  (2026-05-28T19:58:43Z)
• Barriers to Universal Reasoning With Transformers (And How to Overcome Them) [46.59247939156339]
  Chain-of-Thought (CoT) has been shown to empirically improve Transformers' performance, and theoretically increase their expressivity to Turing completeness.<n>We use recent theoretical frameworks for Transformer length generalization and find that -- under standard positional encodings and a finite alphabet -- Transformers with CoT cannot solve problems beyond $TC0$.<n>We attain a length-generalizable simulation of Turing machines where the CoT trace length is linear in the simulated runtime up to a constant.
  arXiv  Detail & Related papers  (2026-04-28T16:10:37Z)
• Generalization Bounds for Transformer Channel Decoders [61.55280736553095]
  This paper studies the generalization performance of ECCT from a learning-theoretic perspective.<n>To the best of our knowledge, this work provides the first theoretical generalization guarantees for this class of decoders.
  arXiv  Detail & Related papers  (2026-01-11T15:56:37Z)
• Quantitative Bounds for Length Generalization in Transformers [58.175107357008876]
  We study the problem of length generalization (LG) in transformers.<n>LG occurs when the internal behavior of the transformer on longer sequences can be "simulated" by its behavior on shorter sequences.
  arXiv  Detail & Related papers  (2025-10-30T21:31:36Z)
• The Role of Sparsity for Length Generalization in Transformers [58.65997625433689]
  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.
  arXiv  Detail & Related papers  (2025-02-24T03:01:03Z)
• 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)
• Sequence Length Independent Norm-Based Generalization Bounds for Transformers [21.2523248114561]
  This paper provides norm-based generalization bounds for the Transformer architecture that do not depend on the input sequence length. We use three novel covering number bounds for the function class of bounded linear transformations to upper bound the Rademacher complexity of the Transformer.
  arXiv  Detail & Related papers  (2023-10-19T18:31:09Z)
• 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)
• Sumformer: Universal Approximation for Efficient Transformers [2.4832703558223725]
  We introduce Sumformer, a novel and simple architecture capable of universally approxingimating sequence-to-sequence functions. We derive a new proof for Transformers, showing that just one attention layer is sufficient for universal approximation.
  arXiv  Detail & Related papers  (2023-07-05T13:59:35Z)
• Combiner: Full Attention Transformer with Sparse Computation Cost [142.10203598824964]
  We propose Combiner, which provides full attention capability in each attention head while maintaining low computation complexity. We show that most sparse attention patterns used in existing sparse transformers are able to inspire the design of such factorization for full attention. An experimental evaluation on both autoregressive and bidirectional sequence tasks demonstrates the effectiveness of this approach.
  arXiv  Detail & Related papers  (2021-07-12T22:43:11Z)
• Length-Adaptive Transformer: Train Once with Length Drop, Use Anytime with Search [84.94597821711808]
  We extend PoWER-BERT (Goyal et al., 2020) and propose Length-Adaptive Transformer that can be used for various inference scenarios after one-shot training. We conduct a multi-objective evolutionary search to find a length configuration that maximizes the accuracy and minimizes the efficiency metric under any given computational budget. We empirically verify the utility of the proposed approach by demonstrating the superior accuracy-efficiency trade-off under various setups.
  arXiv  Detail & Related papers  (2020-10-14T12:28:08Z)

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.