Pocket Foundation Models: Distilling TFMs into CPU-Ready Gradient-Boosted Trees
For practitioners needing fast, CPU-based tabular inference, this provides a practical distillation pipeline that preserves most of the TFM's accuracy while drastically reducing latency.
The authors distill tabular foundation models (TFMs) into gradient-boosted trees (XGBoost/CatBoost) that run on CPU in under 2 ms, achieving 96.5% of teacher AUC (0.882 macro-mean) across 153 datasets, with a 38x to 860x speedup over GPU-based TFMs. They solve the label leakage problem in in-context learning teachers using stratified out-of-fold labeling.
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.