google/tabfm-1.0.0-pytorch · Hugging Face

Log In<br>Sign Up

TabFM 1.0.0 (PyTorch)

TabFM is a zero-shot tabular foundation model from Google Research. It supports<br>classification and regression on structured/tabular data with mixed numerical and<br>categorical columns, requiring no fine-tuning or hyperparameter search - training<br>examples are passed as context and predictions are made in a single forward pass.

This repository contains the PyTorch weights. For the JAX/Flax weights see<br>google/tabfm-1.0.0-jax.

Getting Started

pip install tabfm[pytorch]

Classification:

from tabfm import TabFMClassifier, tabfm_v1_0_0_pytorch as tabfm_v1_0_0

model = tabfm_v1_0_0.load(model_type="classification")<br>clf = TabFMClassifier(model=model)<br>clf.fit(X_train, y_train)<br>probs = clf.predict_proba(X_test)

Regression:

from tabfm import TabFMRegressor, tabfm_v1_0_0_pytorch as tabfm_v1_0_0

model = tabfm_v1_0_0.load(model_type="regression")<br>reg = TabFMRegressor(model=model)<br>reg.fit(X_train, y_train)<br>preds = reg.predict(X_test)

You can also load directly using the HuggingFace Hub API:

from tabfm.src.pytorch.tabfm_v1_0_0 import TabFM_HF

clf_model = TabFM_HF.from_pretrained("google/tabfm-1.0.0-pytorch", subfolder="classification")<br>reg_model = TabFM_HF.from_pretrained("google/tabfm-1.0.0-pytorch", subfolder="regression")

Available Checkpoints

Subfolder<br>Task<br>is_classifier

classification/<br>Classification (up to 10 classes)<br>True

regression/<br>Regression<br>False

Developers and Affiliations

Developed by the Google Research team.

Intended Use

Tabular data with numerical and/or categorical columns

Binary and multiclass classification (up to 10 classes)

Regression on continuous targets

Zero-shot inference: no dataset-specific training or hyperparameter tuning

Works with DataFrames (pandas) or numpy arrays

Not Intended For

Images, audio, video, or raw text

More than 10 output classes (hard model limit)

Tasks requiring task-specific fine-tuning

Non-tabular structured data (graphs, sequences)

Commercial use (see License below)

Model Architecture

TabFM uses alternating row and column attention to capture both feature interactions<br>and row-level patterns:

Column attention (Set Transformer): embeds each cell using Fourier features and<br>a per-group linear projection, then aggregates across rows via induced self-attention

Row compression : CLS tokens summarise each row into a dense vector via row-level<br>attention with Rotary Position Embedding (RoPE)

ICL Transformer : a 24-block causal transformer operates over the compressed row<br>vectors, treating training rows as context and outputting predictions for test rows

Key hyperparameters:

Parameter<br>Value

Embedding dim<br>256

Column attention blocks<br>3 (4 heads, 256 induced points)

Row attention blocks<br>3 (8 heads, 8 CLS tokens)

ICL transformer blocks<br>24 (8 heads)

Feed-forward factor

Max classes<br>10

Activation<br>SwiGLU

Fourier features<br>32 frequencies

Training Data and Priors

TabFM was trained on hundreds of millions of synthetic datasets generated<br>dynamically using structural causal models (SCMs). Synthetic data was chosen due to<br>the scarcity of diverse, high-quality open-source tabular datasets and to avoid<br>privacy/licensing concerns with real-world industrial data. The SCM prior encodes<br>inductive biases about causal structure and feature relationships typical in tabular<br>tasks.

Performance

TabFM was evaluated on TabArena across 51 datasets<br>(38 classification, 13 regression). In zero-shot mode - a single forward pass with no<br>hyperparameter search - TabFM outperforms heavily-tuned supervised baselines including<br>gradient-boosted trees. The TabFMClassifier.ensemble() preset (feature crosses,<br>SVD features, NNLS blending) yields further improvements.

See the Google Research blog post for full benchmark details.

Ethical Considerations

TabFM was trained entirely on synthetic data. Performance on specific real-world<br>domains, minority groups, or edge distributions is not fully characterised. Users<br>should evaluate the model on held-out data representative of their use case before<br>deploying in high-stakes settings.

Limitations

Max 10 classes for classification (hard architectural limit)

Memory usage scales with the number of training rows (all rows are passed as context)

Optimised for tables up to 500 features; behaviour on very wide tables may degrade

Performance is not guaranteed to match task-specific, fine-tuned models on all datasets

Not an officially supported Google product

License

The model weights in this repository are released under the<br>TabFM Non-Commercial License v1.0 - see LICENSE. The source code is<br>Apache 2.0 licensed via google-research/tabfm.

Version

1.0.0

Citation

@article{tabfm2026,<br>title = {TabFM: A Zero-Shot Foundation Model for Tabular Data},<br>author = {Google Research},<br>year = {2026},<br>url = {https://research.google/blog/introducing-tabfm-a-zero-shot-foundation-model-for-tabular-data/}

Downloads last month 1,177

Inference Providers NEW

Tabular...