TabPFN

What it is

TabPFN is a pretrained transformer that predicts on small to medium tabular datasets with one pass through the network, with no training or hyperparameter tuning needed for each new dataset. The Python package exposes scikit-learn style classifier and regressor classes, downloads model checkpoints on first use, and runs on Nvidia, AMD, or Apple Silicon GPUs, or on CPU for smaller datasets. Data scientists use it as a quick, strong baseline for spreadsheet-style prediction tasks. Recent model weights carry non-commercial licenses, while the older v2 weights are Apache-based.