The Illusion of Generalization in Tabular Language Models

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Gorla, Aditya, Puduppully, Ratish
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866913171642515456
author Gorla, Aditya
Puduppully, Ratish
author_facet Gorla, Aditya
Puduppully, Ratish
contents Tabular Language Models (TLMs) have been claimed to achieve strong generalization for tabular prediction. We conduct a systematic re-evaluation of Tabula-8B as a representative TLM, utilizing 165 datasets from the UniPredict benchmark. Our investigation reveals three findings. First, binary and categorical classification achieve near-zero median lift over majority-class baselines and strong aggregate performance is driven entirely by quartile classification tasks. Second, top-performing datasets exhibit pervasive contamination, including complete train-test overlap and task-level leakage that evades standard deduplication. Third, instruction-tuning without tabular exposure recovers 92.2% of standard classification performance and on quartile classification, format familiarity closes 71.3% of the gap with the residual attributable to contaminated datasets. These findings suggest claimed generalization likely reflects evaluation artifacts rather than learned tabular reasoning. We conclude with recommendations for strengthening TLM evaluation.
format Preprint
id arxiv_https___arxiv_org_abs_2602_04031
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle The Illusion of Generalization in Tabular Language Models
Gorla, Aditya
Puduppully, Ratish
Machine Learning
Tabular Language Models (TLMs) have been claimed to achieve strong generalization for tabular prediction. We conduct a systematic re-evaluation of Tabula-8B as a representative TLM, utilizing 165 datasets from the UniPredict benchmark. Our investigation reveals three findings. First, binary and categorical classification achieve near-zero median lift over majority-class baselines and strong aggregate performance is driven entirely by quartile classification tasks. Second, top-performing datasets exhibit pervasive contamination, including complete train-test overlap and task-level leakage that evades standard deduplication. Third, instruction-tuning without tabular exposure recovers 92.2% of standard classification performance and on quartile classification, format familiarity closes 71.3% of the gap with the residual attributable to contaminated datasets. These findings suggest claimed generalization likely reflects evaluation artifacts rather than learned tabular reasoning. We conclude with recommendations for strengthening TLM evaluation.
title The Illusion of Generalization in Tabular Language Models
topic Machine Learning
url https://arxiv.org/abs/2602.04031