Is One Layer Enough? Understanding Inference Dynamics in Tabular Foundation Models

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Balef, Amir Rezaei, Koshil, Mykhailo, Eggensperger, Katharina
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866909022611832832
author Balef, Amir Rezaei
Koshil, Mykhailo
Eggensperger, Katharina
author_facet Balef, Amir Rezaei
Koshil, Mykhailo
Eggensperger, Katharina
contents Transformer-based tabular foundation models (TFMs) dominate small to medium tabular predictive benchmark tasks, yet their inference mechanisms remain largely unexplored. We present the first large-scale mechanistic study of layerwise dynamics in 6 state-of-the-art tabular in-context learning models. We explore how predictions emerge across depth, identify distinct stages of inference and reveal latent-space dynamics that differ from those of language models. Our findings indicate substantial depthwise redundancy across multiple models, suggesting iterative refinement with overlapping computations during inference stages. Guided by these insights, we design a proof-of-concept, looped single-layer model that uses only 20% of the original model's parameters while achieving comparable performance. The code is available at https://github.com/amirbalef/is_one_layer_enough.
format Preprint
id arxiv_https___arxiv_org_abs_2605_06510
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Is One Layer Enough? Understanding Inference Dynamics in Tabular Foundation Models
Balef, Amir Rezaei
Koshil, Mykhailo
Eggensperger, Katharina
Machine Learning
Artificial Intelligence
Transformer-based tabular foundation models (TFMs) dominate small to medium tabular predictive benchmark tasks, yet their inference mechanisms remain largely unexplored. We present the first large-scale mechanistic study of layerwise dynamics in 6 state-of-the-art tabular in-context learning models. We explore how predictions emerge across depth, identify distinct stages of inference and reveal latent-space dynamics that differ from those of language models. Our findings indicate substantial depthwise redundancy across multiple models, suggesting iterative refinement with overlapping computations during inference stages. Guided by these insights, we design a proof-of-concept, looped single-layer model that uses only 20% of the original model's parameters while achieving comparable performance. The code is available at https://github.com/amirbalef/is_one_layer_enough.
title Is One Layer Enough? Understanding Inference Dynamics in Tabular Foundation Models
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2605.06510