# google-research/tabfm

TabFM (Tabular Foundation Model) is a pretrained tabular foundation model developed by Google Research for tabular data regression and classification.

Repository: https://github.com/google-research/tabfm
Canonical: https://ross.abutalabs.com/products/tabfm
Homepage: https://research.google/blog/introducing-tabfm-a-zero-shot-foundation-model-for-tabular-data/
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-08-18T17:12:51+00:00

## Health v2 (maintenance only)
Score: 67/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 98, release rhythm 62, longevity 5
- inputs: {"age_days": 78, "days_push": 15, "days_rel": 43, "gap_med": null, "n_releases_24m": 1}
- flags: young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2551, forks 264 (observed 2026-08-28T04:07:00.510612+00:00)

## What it is
TabFM is a scikit-learn compatible tabular foundation model from Google Research that performs zero-shot classification and regression on tabular datasets with mixed column types. It uses in-context learning at inference time, requiring no parameter training on the user's dataset, with JAX and PyTorch backends.

## Use cases
- zero-shot classification on tabular data
- zero-shot regression on tabular datasets
- predict customer churn without training a model
- replace XGBoost hyperparameter tuning with instant predictions
- classify mixed numerical and categorical features
- quickly benchmark tabular models without feature engineering

## When to choose
- you need fast predictions on new tabular datasets without training or hyperparameter tuning
- your data mixes numerical and categorical columns
- you want a scikit-learn compatible drop-in estimator
- non-commercial use of pretrained weights is acceptable

## When to avoid
- you need a commercially licensed model — the default pretrained weights are non-commercial only
- you have large datasets where gradient-boosted trees like XGBoost already perform well
- you need full control over model training and fine-tuning
- you require a lightweight dependency-free solution

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, llm-inference
- domain: machine-learning, data-science, large-language-models
- platform: python, cross-platform
- tags: tabular-data, foundation-model, zero-shot, in-context-learning, scikit-learn-compatible, jax, pytorch, classification, regression

## Member repositories
- google-research/tabfm (main) score 67

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:00.510612+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:23:53.879871+00:00, confidence not recorded.
  - readme: https://github.com/google-research/tabfm (fetched 2026-08-28T04:07:00.510612+00:00, sha abae9c62df36)
  - homepage: https://research.google/blog/introducing-tabfm-a-zero-shot-foundation-model-for-tabular-data/ (fetched 2026-08-29T10:06:28.833836+00:00, sha dc5ef793fcd9)
  - registry_pypi: https://pypi.org/pypi/tabfm/json (fetched 2026-08-29T10:06:28.843124+00:00, sha c1cac915ce81)
- Data as of 2026-08-30T08:39:29.467469+00:00.