Set-LLM: A Permutation-Invariant LLM

Beni Egressy, Jan Stühmer · 2025

Read Set-LLM: A Permutation-Invariant LLM by Beni Egressy, Jan Stühmer (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-15433.txt.

Description

Beni Egressy and Jan Stuhmer present Set-LLM, a change to large language models that makes their answers steady under reordering, in a 2025 paper. Models often favor the first option or change answers when options are shuffled. That bias harms uses from multiple-choice tests to automated judging of model outputs.

Their design lets a model take mixed set and text inputs with a guarantee that order does not matter. It adds a new attention mask and new position codes made for sets. They prove the invariance in theory and show it in tests. The work targets a clear robustness flaw in current models.

Catalog record

CREATORBeni Egressy, Jan Stühmer
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE53 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-15433

Browse the full library · About the Tower of Babel