FairTFM is a
tabular foundation model for fairness-aware, in-context prediction on tabular data. It is a transformer-based architecture (built on components from
nanoTabPFN) that jointly embeds features, targets, and a
sensitive attribute via a dedicated
SensitiveAttributeEncoder, then predicts labels for unseen (test) rows conditioned on a small set of in-context training rows — no per-dataset fine-tuning required.
Four checkpoints are provided, corresponding to different fairness-regularization strengths (λ) used to trace a fairness/accuracy Pareto front: λ = 0.7, 1.0, 10, 25. Higher λ trades predictive performance for lower fairness-metric disparity.
1from fairtfm import FairTFMClassifier, compute_fairness_metrics
2
3# Load checkpoint
4classifier = FairTFMClassifier(model="path/to/checkpoint")
5
6# Fit on training data
7classifier.fit(X_train, y_train, s_train)
8
9# Predict
10predictions = classifier.predict(X_test, s_test)
11probabilities = classifier.predict_proba(X_test, s_test)
12
13# Fairness metrics (returns dict with performance metrics)
14compute_fairness_metrics(X_test, y_test, s_test)
1@inproceedings{
2kenfack2026training,
3title={Training Fair Tabular Foundation Models},
4author={Patrik Kenfack and Jesse C. Cresswell and Anthony L. Caterini and Samira Ebrahimi Kahou and Ulrich A{\"\i}vodji},
5booktitle={2nd ICML Workshop on Foundation Models for Structured Data},
6year={2026},
7url={https://openreview.net/forum?id=ajIvCEbadL}
8}
