REaLTabFormer
github.com
- Free plan
- Windows
- macOS
- Linux
- Self-hosted
What REaLTabFormer is
REaLTabFormer is a free, MIT-licensed framework for synthesizing tabular and relational data. Its relational approach uses a sequence-to-sequence model, while its model for non-relational tables uses GPT-2 and supports independent observations. Examples use pandas DataFrames as input; relational generation requires matching join-key columns in the parent and child tables. The documented workflow fits a model, saves it locally, and samples synthetic data. For non-relational training, the model stops when the synthetic distribution is close to the real distribution. The framework also offers observation validators to filter invalid samples, including a GeoValidator example. The paper describes target masking to prevent data copying and the Qδ statistic with statistical bootstrapping to detect overfitting. Install the package from PyPI with pip install realtabformer; the current package requires Python 3.8 or newer and is classified as operating-system independent. Linux, macOS, Windows, and self-hosted use are listed. The project describes use in research or projects and asks users to cite its paper.
Who it suits
REaLTabFormer suits researchers and project teams generating synthetic tabular or relational datasets. Users should be comfortable working with Python and, for relational data, matching join-key columns.
Good to know, and what to watch
- Supports relational and non-relational synthetic data generation.
- Observation validators can filter invalid samples.
- Includes privacy-oriented design measures described in the paper.
- Free and distributed under the MIT License.
- Classified as operating-system independent.
- The current PyPI package requires Python 3.8 or newer.
- Relational generation requires matching join-key columns.
- The documented workflow involves fitting and saving a model locally.
Plans
Every plan and priceREaLTabFormer
Open source
The verdict
REaLTabFormer provides distinct approaches for tabular and relational synthetic data, plus validation tools. Its Python requirement and join-key needs are important setup considerations.
What the maker publishes
- Data validation
- The framework provides an interface for observation validators that filter invalid synthetic samples, including a GeoValidator example. source
- Development context
- The project acknowledges funding from the World Bank-UNHCR Joint Data Center on Forced Displacement for work involving responsible microdata access and synthetic population research. source
- Documented audience
- The project describes its use for projects or research and asks users to cite its research paper when using it. source
- Funding
- The project acknowledges funding from the World Bank-UNHCR Joint Data Center on Forced Displacement. source
- Input format
- Examples use pandas DataFrames as model input. source
- Installation
- The package is installed from PyPI with pip install realtabformer. source
- Installation
- The project is available through PyPI and documents installation with pip for Python 3.7 or later. source
- License
- The package is distributed under the MIT License. source
- License
- The repository provides the software under the MIT License, which permits use, modification, distribution, sublicensing, and sale subject to its terms. source
- Operating systems
- PyPI classifies the package as operating-system independent. source
- Privacy-oriented design
- The paper says target masking is used to prevent data copying and the Qδ statistic with statistical bootstrapping is used to detect overfitting. source
- Purpose
- REaLTabFormer is a unified framework for synthesizing different types of tabular data. source
- Purpose
- REaLTabFormer is a framework for generating synthetic tabular and relational data with transformer models. source
- Python requirement
- The current PyPI package requires Python 3.8 or newer. source
- Relational generation
- It uses a sequence-to-sequence model to generate synthetic relational datasets. source
- Relational keys
- Relational generation requires matching join-key columns in the parent and child tables. source
- Relational model
- A sequence-to-sequence model generates synthetic relational datasets. source
- Release
- PyPI lists version 0.2.4 as released on January 4, 2026. source
- Sampling
- The documented workflow fits a model, saves it locally, and samples synthetic data from it. source
- Security reporting
- The security policy asks users to report vulnerabilities by email rather than through public GitHub issues and says a response should arrive within 48 hours. source
- Stopping criterion
- For non-relational tabular training, the model stops when the synthetic distribution is close to the real distribution. source
- Support
- For vulnerability reports, the policy lists [email protected] and requests details that help reproduce and assess the issue. source
- Tabular model
- Its non-relational tabular model uses GPT-2 and can model tabular data with independent observations out of the box. source
- Tabular model
- For independent tabular observations, it uses GPT-2 and can model data out of the box. source
Checked against the maker's pages on 2 Oct 2026.