# Pocket Foundation Models: Distilling TFMs into CPU-Ready Gradient-Boosted Trees

arXiv: https://arxiv.org/abs/2605.18654 · HTML: https://mlsanity.com/paper/2605.18654/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in General knowledge distillation methods, one of 12 directions in Knowledge distillation.
- Builds on 20 of its 32 references that are in the map.
- First posted on arXiv 18 May 2026.

## Placed in

- Knowledge distillation / [General knowledge distillation methods](https://mlsanity.com/distill/general-knowledge-distillation-methods/)

## Builds on (most cited first, 20 of 20)

- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [XGBoost: A Scalable Tree Boosting System](https://mlsanity.com/paper/1603.02754/) (2016)
- [On Calibration of Modern Neural Networks](https://mlsanity.com/paper/1706.04599/) (2017)
- [TabPFN: A Transformer That Solves Small Tabular Classification Problems in a Second](https://mlsanity.com/paper/2207.01848/) (2022)
- [TabNet: Attentive Interpretable Tabular Learning](https://mlsanity.com/paper/1908.07442/) (2019)
- [Born Again Neural Networks](https://mlsanity.com/paper/1805.04770/) (2018)
- [Do Deep Nets Really Need to be Deep?](https://mlsanity.com/paper/1312.6184/) (2013)
- [When Do Neural Nets Outperform Boosted Trees on Tabular Data?](https://mlsanity.com/paper/2305.02997/) (2023)
- [Why do tree-based models still outperform deep learning on tabular data?](https://mlsanity.com/paper/2207.08815/) (2022)
- [OpenML Benchmarking Suites](https://mlsanity.com/paper/1708.03731/) (2017)
- [TabArena: A Living Benchmark for Machine Learning on Tabular Data](https://mlsanity.com/paper/2506.16791/) (2025)
- [TabICLv2: A better, faster, scalable, and open tabular foundation model](https://mlsanity.com/paper/2602.11139/) (2026)
- [TabPFN-2.5: Advancing the State of the Art in Tabular Foundation Models](https://mlsanity.com/paper/2511.08667/) (2025)
- [TabDPT: Scaling Tabular Foundation Models on Real Data](https://mlsanity.com/paper/2410.18164/) (2024)
- [LimiX: Unleashing Structured-Data Modeling Capability for Generalist Intelligence](https://mlsanity.com/paper/2509.03505/) (2025)
- [Orion-MSP: Multi-Scale Sparse Attention for Tabular In-Context Learning](https://mlsanity.com/paper/2511.02818/) (2025)
- [TabTune: A Unified Library for Inference and Fine-Tuning Tabular Foundation Models](https://mlsanity.com/paper/2511.02802/) (2025)
- [Dataset distillation for memorized data: Soft labels can leak held-out teacher knowledge](https://mlsanity.com/paper/2506.14457/) (2025)
- [Orion-Bix: Bi-Axial Attention for Tabular In-Context Learning](https://mlsanity.com/paper/2512.00181/) (2025)
- [Data Laundering: Artificially Boosting Benchmark Results through Knowledge Distillation](https://mlsanity.com/paper/2412.15255/) (2024)

## Built on by (most cited first, 2 of 2)

- [GEAR: Generative Expansion and Real Anchoring for Two-Stage Distillation of Tabular Foundation Models](https://mlsanity.com/paper/2608.18849/) (2026)
- [Distillation of Tabular Foundation Models into Efficient Predictors](https://mlsanity.com/paper/2610.01435/) (2026)

## Abstract (arXiv)

A fraud scorer needs to answer in under 2 ms. The best tabular foundation models (TFMs) take 151-1,275 ms on GPU. We close this gap by distilling the TFM offline into an XGBoost or CatBoost student that runs natively on CPU. The central obstacle is specific to in-context learning (ICL) teachers: they leak labels when scoring their own training set, so the soft targets collapse to near-one-hot vectors with no inter-class structure left to distill. Stratified out-of-fold (OOF) teacher labeling prevents this. Across 153 classification datasets drawn from TALENT, OpenML-CC18, TabZilla, and TabArena, distilling TabICLv2 into XGBoost gives 0.882 macro-mean AUC (96.5% of teacher AUC) at 1.9 ms on CPU, a 38x to 860x speedup across teacher-student pairs with a statistically significant edge over a tuned CatBoost baseline (Wilcoxon p = 0.0008; 51% win rate). Four further findings: teacher rank transfers exactly to student rank; gains concentrate on low-dimensional data (< 21 features: +0.011 over CatBoost vs. >21 features: +0.001); multi-teacher averaging helps MLP students (+0.006, p = 0.003) but adds less than 0.001 for tree students; and on high-dimensional tasks where the teacher itself trails CatBoost, distillation makes things worse rather than better. The full pipeline is open-sourced as part of the TabTune library.