Turning Tabular Foundation Models into Graph Foundation Models

This is the official repository for the paper "Turning Tabular Foundation Models into Graph Foundation Models" (arXiv). In this repository, we provide code for reproducing our key experiments.

Note

See also: GraphPFN, our next step toward applying TFMs and PFNs to graph node-level tasks. GraphPFN augments the LimiX tabular foundation model with graph adapters and is pretrained on millions of synthetic datasets, achieving state-of-the-art results.

Reproducing Experiments

Prerequisites

1. Install uv
2. Install dependencies

uv sync

3. Download TabPFNv2 checkpoints

wget https://huggingface.co/Prior-Labs/TabPFN-v2-reg/resolve/main/tabpfn-v2-regressor.ckpt?download=true -O checkpoints/tabpfn-v2-regressor.ckpt
wget https://huggingface.co/Prior-Labs/TabPFN-v2-clf/resolve/main/tabpfn-v2-classifier.ckpt?download=true -O checkpoints/tabpfn-v2-classifier.ckpt

4. For experiments on GraphLand, download datasets and place them in "data" directory

Running the code

You can execute a minimal run (G2T-LimiX finetuning with 10 ensemble members) with the following command:

uv run bin/go.py exp/g2t/main/limix/finetune/10/tolokers-2/tuning.toml --force

Project Structure

• bin/ - Training and evaluation scripts
• exp/ - Experiment configurations and results
• lib/ - Common utilities and tools
• vendor/ – Vendored third-party code with minor import modifications for compatibility

Configuration

Experiments are configured using TOML files located in the exp/ directory. Each configuration specifies:

• Dataset path and preprocessing
• Model hyperparameters
• Training settings
• Evaluation metrics

Results

After training, results are saved in the same directory as the configuration file:

• report.json - Evaluation metrics
• Model checkpoints
• Training logs

Licenses

This project uses third-party components LimiX, TabICL and TabPFN. See the NOTICE file and LICENSES/ directory for details.

---

Built with PriorLabs-TabPFN