Contamination Map Methodology

Tabular / Model

Real-TabPFN

Organisation
Prior Labs
Family
TabPFN · version v2 (real-data continued pre-training)
Released
2025-07-05 (arXiv v1 submission date https://arxiv.org/abs/2507.03971)
Weights
Open
Links
PaperCode

Training data disclosure: Full list

Level determined for 5% of benchmark cells.

“Appendix A lists the curated datasets used for continued pre-training.”
arxiv.org · paper · Section 2 · retrieved 2026-10-15
“The following table lists the 71 datasets curated for continued pre-training, along with their source and access link.”
arxiv.org · paper · Appendix A · retrieved 2026-10-15

Other declared training data

Benchmark cells

TabArena

Unknown airfoil_self_noise · training data not disclosed in enough detail

Seen, split unclear Amazon_employee_access · the dataset was used; the source does not say which part

Evidence (1)
“Amazon employee access”
arxiv.org · paper · Appendix A, Training Datasets table (source: OpenML) · retrieved 2026-10-15

Unknown anneal · training data not disclosed in enough detail

Unknown Another-Dataset-on-used-Fiat-500 · training data not disclosed in enough detail

Unknown APSFailure · training data not disclosed in enough detail

Seen, split unclear bank-marketing · the dataset was used; the source does not say which part

Evidence (1)
“Bank marketing data set UCI”
arxiv.org · paper · Appendix A, Training Datasets table (source: OpenML) · retrieved 2026-10-15

Unknown Bank_Customer_Churn · training data not disclosed in enough detail

Unknown Bioresponse · training data not disclosed in enough detail

Unknown blood-transfusion-service-center · training data not disclosed in enough detail

Unknown churn · training data not disclosed in enough detail

Unknown coil2000_insurance_policies · training data not disclosed in enough detail

Unknown concrete_compressive_strength · training data not disclosed in enough detail

Unknown credit-g · training data not disclosed in enough detail

Unknown credit_card_clients_default · training data not disclosed in enough detail

Unknown customer_satisfaction_in_airline · training data not disclosed in enough detail

Unknown diabetes · training data not disclosed in enough detail

Unknown Diabetes130US · training data not disclosed in enough detail

Unknown diamonds · training data not disclosed in enough detail

Unknown E-CommereShippingData · training data not disclosed in enough detail

Unknown Fitness_Club · training data not disclosed in enough detail

Unknown Food_Delivery_Time · training data not disclosed in enough detail

Unknown GiveMeSomeCredit · training data not disclosed in enough detail

Unknown hazelnut-spread-contaminant-detection · training data not disclosed in enough detail

Unknown healthcare_insurance_expenses · training data not disclosed in enough detail

Unknown heloc · training data not disclosed in enough detail

Unknown hiva_agnostic · training data not disclosed in enough detail

Unknown houses · training data not disclosed in enough detail

Unknown HR_Analytics_Job_Change_of_Data_Scientists · training data not disclosed in enough detail

Unknown in_vehicle_coupon_recommendation · training data not disclosed in enough detail

Unknown Is-this-a-good-customer · training data not disclosed in enough detail

Unknown kddcup09_appetency · training data not disclosed in enough detail

Unknown Marketing_Campaign · training data not disclosed in enough detail

Unknown maternal_health_risk · training data not disclosed in enough detail

Unknown miami_housing · training data not disclosed in enough detail

Unknown MIC · training data not disclosed in enough detail

Unknown NATICUSdroid · training data not disclosed in enough detail

Seen, split unclear online_shoppers_intention · the dataset was used; the source does not say which part

Evidence (1)
“Online Shopper’s Intention”
arxiv.org · paper · Appendix A, Training Datasets table (source: Kaggle) · retrieved 2026-10-15

Unknown physiochemical_protein · training data not disclosed in enough detail

Unknown polish_companies_bankruptcy · training data not disclosed in enough detail

Unknown qsar-biodeg · training data not disclosed in enough detail

Unknown QSAR-TID-11 · training data not disclosed in enough detail

Unknown QSAR_fish_toxicity · training data not disclosed in enough detail

Unknown SDSS17 · training data not disclosed in enough detail

Unknown seismic-bumps · training data not disclosed in enough detail

Unknown splice · training data not disclosed in enough detail

Unknown students_dropout_and_academic_success · training data not disclosed in enough detail

Unknown superconductivity · training data not disclosed in enough detail

Unknown taiwanese_bankruptcy_prediction · training data not disclosed in enough detail

Unknown website_phishing · training data not disclosed in enough detail

Unknown wine_quality · training data not disclosed in enough detail

Unknown jm1 · training data not disclosed in enough detail

Notes

TabPFN v2 further pre-trained on 71 real datasets from OpenML and Kaggle. Deduplication was only against its own evaluation suite (29 AMLB datasets with <10,000 samples), not against TabArena: 'We exclude any dataset from the pre-training data that does not meet these criteria.' The paper says the weights are public on Hugging Face but does not name the checkpoint file. Several training datasets match TabArena datasets by name (Amazon employee access, Online Shopper's Intention, Bank marketing).

Sources