Saved in:
Bibliographic Details
Main Author: Sun, Simeng
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2509.25073
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915522432466944
author Sun, Simeng
author_facet Sun, Simeng
contents We study Transformers on the task \emph{program trace generation} (PTG), where models produce step-by-step execution traces for synthetic programs. Unlike existing algorithmic problems, PTG externalizes reasoning through long traces where each step is trivial. We train small Transformers with diverse modifications, including alternative position encodings, softmax replacements, hybrid model, and short convolutions. While these models achieve strong in-distribution accuracy, they exhibit systematic failures when generalizing to various factors (e.g., program length, trace steps), though some designs significantly improve generalization.
format Preprint
id arxiv_https___arxiv_org_abs_2509_25073
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle An empirical study on the limitation of Transformers in program trace generation
Sun, Simeng
Computation and Language
We study Transformers on the task \emph{program trace generation} (PTG), where models produce step-by-step execution traces for synthetic programs. Unlike existing algorithmic problems, PTG externalizes reasoning through long traces where each step is trivial. We train small Transformers with diverse modifications, including alternative position encodings, softmax replacements, hybrid model, and short convolutions. While these models achieve strong in-distribution accuracy, they exhibit systematic failures when generalizing to various factors (e.g., program length, trace steps), though some designs significantly improve generalization.
title An empirical study on the limitation of Transformers in program trace generation
topic Computation and Language
url https://arxiv.org/abs/2509.25073