analiza aiqbench
This study provides a mechanistic analysis of weight-tied transformers, demonstrating that iterative looping enables models to solve depth-bound algorithmic tasks by gradually advancing through intermediate nodes rather than merely simulating deeper stacks. The research establishes that beyond a minimum effective depth, the degree of weight tying becomes the critical factor for success, offering new insights into how architectures can efficiently learn iterative programs without increasing parameter counts or inference costs.
Inside the Loop: A Mechanistic Study of Weight-Tied Transformers on Depth-Bound Algorithmic Tasks
arXiv:2610.24135v1 Announce Type: new Abstract: Solving algorithmic tasks with many sequential steps typically requires scaling up transformer depth or chain-of-thought length, incurring unfavorable parameter and inference-compute costs, respectively. Looped transformers, which reuse one block of layers across many iterations, are poised to address this, since they scale effective depth without adding parameters or tokens. Yet, it remains unclear what each iteration is actually computing, or whether the loop merely simulates a deeper untied stack. Here, we study directed permutation traversal, in which a model follows $H$ hops from a starting node and reports the node it reaches, so every intermediate hop has a known ground truth. We find that a minimum effective depth is necessary to solve the task, beyond which the degree of weight tying becomes the determining factor. For instance, a looped transformer with a four-layer block iterated eight times solves the task in every seed, while an identically-trained untied transformer of the same effective depth never does, succeeding only with a lower learning rate or supervision of its intermediate layers. Probing and intervening on the residual stream, we find that the looped model advances through the intermediate nodes gradually, forming a staircase that adapts to its iteration budget, whereas the untied model, at that lower rate, exposes them only in its final layers. Taken together, our results recast looped transformers as iterative program learners, which allocate their iterations to successive steps of the algorithm and solve tasks that identically-trained untied models of the same effective depth do not. Code is available at https://github.com/magnus-sesodia/insidetheloop.
Abstract:Solving algorithmic tasks with many sequential steps typically requires scaling up transformer depth or chain-of-thought length, incurring unfavorable parameter and inference-compute costs, respectively. Looped transformers, which reuse one block of layers across many iterations, are poised to address this, since they scale effective depth without adding parameters or tokens. Yet, it remains unclear what each iteration is actually computing, or whether the loop merely simulates a deeper untied stack. Here, we study directed permutation traversal, in which a model follows $H$ hops from a starting node and reports the node it reaches, so every intermediate hop has a known ground truth. We find that a minimum effective depth is necessary to solve the task, beyond which the degree of weight tying becomes the determining factor. For instance, a looped transformer with a four-layer block iterated eight times solves the task in every seed, while an identically-trained untied transformer of the same effective depth never does, succeeding only with a lower learning rate or supervision of its intermediate layers. Probing and intervening on the residual stream, we find that the looped model advances through the intermediate nodes gradually, forming a staircase that adapts to its iteration budget, whereas the untied model, at that lower rate, exposes them only in its final layers. Taken together, our results recast looped transformers as iterative program learners, which allocate their iterations to successive steps of the algorithm and solve tasks that identically-trained untied models of the same effective depth do not. Code is available at this https URL.
Accepted to NeurIPS 2026. 17 pages, 12 figures
Machine Learning (cs.LG); Computation and Language (cs.CL)
arXiv:2610.24135 [cs.LG]
(or arXiv:2610.24135v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.24135 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
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 · 359 cuvinte
Subiecte conexe
Transformer Field Theory: A Response-Theoretic Approach to Mechanistic Interpretability
arXiv · 12 iun. 2026
Ablation-Reversible Heads Don't Transfer: A Stress Test for Mechanistic Role Claims in Transformers
arXiv · 9 iun. 2026
SpIn-ViT: Designing a Sparsity-Induced Vision Transformer That Is Mechanistically Interpretable
arXiv · 18 aug. 2026
Conditional Co-Ablation: Recovering Self-Repair Backups in Transformer Circuits
arXiv · 23 sept. 2026