tabpfn

What it is and what it does

TabPFN is a pre-trained foundation model designed to make tabular machine learning faster by eliminating hyperparameter tuning. It provides TabPFNClassifier and TabPFNRegressor classes that follow scikit-learn conventions, so you can fit and predict on structured data with minimal setup. The model downloads weights from huggingface-hub on first use and runs inference via torch, supporting both GPU and CPU execution (though GPU is strongly recommended).

The package depends on torch, scikit-learn, pandas, numpy, scipy, einops, pydantic, joblib, tqdm, and lightgbm. It supports Python 3.10 through 3.14. By default it uses TabPFN-3 (non-commercial license), but you can instantiate earlier versions like TabPFN-2.6 or TabPFN-2 via ModelVersion constants. For production use, Prior Labs offers a commercial Enterprise Edition.

Use it for:

• Rapid prototyping on tabular datasets when you need a strong baseline without manual hyperparameter search.
• Binary and multiclass classification on structured data with minimal preprocessing or model selection overhead.
• Regression tasks on tabular data where you want pre-trained representations without training from scratch.
• Exploratory analysis where you want to quickly assess model performance before investing in tuning.
• Scenarios where you have a GPU available and want to leverage foundation model capabilities on tables.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

TabPFN is a foundation model for tabular data that provides pre-trained classifiers and regressors for rapid inference on structured datasets without requiring hyperparameter tuning.

Yes, with conditions. TabPFN is worth installing if you work with tabular data and have a GPU available, or can tolerate CPU-only inference on moderate datasets. The package is actively maintained with low install friction and genuine value for rapid prototyping. Verify the non-commercial license restrictions for TabPFN-3 (the default) if your use is commercial; if so, you'll need a commercial license or use TabPFN-2 weights (Apache 2.0 with attribution). The attribution requirement is substantial—ensure compliance before distributing.

Install

tabpfn on PyPI

pip install tabpfn

uv add tabpfn

poetry add tabpfn

Installing tabpfn

Before you install

Installation is straightforward with low friction—the package ships as a wheel and handles PyTorch setup automatically. The project is actively maintained with a recent release (1 day old) and strong community engagement (7795 stars). However, optimal use requires a GPU; CPU-only inference is limited to moderate datasets.

License in practice

The code and TabPFN-2 weights use Prior Labs License (Apache 2.0 with an added attribution requirement). TabPFN-3, the default model, uses a non-commercial license. Section 10 requires prominent attribution ("Built with PriorLabs-TabPFN") on websites, UIs, and documentation if you distribute or make the work available. Internal testing without external communication is exempt.

Quickstart

pip install tabpfn

from tabpfn import TabPFNClassifier

clf = TabPFNClassifier()
clf.fit(X_train, y_train)
predictions = clf.predict(X_test)

Requires Python 3.10+. GPU recommended for performance; CPU-only inference limited to moderate datasets.

Verify before relying

• Whether the non-commercial license for TabPFN-3 model weights permits use in commercial products.
• Whether the attribution requirement in Section 10 applies to internal enterprise deployments.
• Performance characteristics and accuracy benchmarks on specific dataset types or sizes.

Package facts

Evidence: tabpfn-8.3.0-py3-none-any.whl

Tags

More Software Development packages