1 minute to HyperBench¶

This page is a quick, “copy/paste and run” introduction to HyperBench.

Prerequisites¶

• Follow the installation guide.
• Recommended tools: uv + make.

Run your first example¶

From the repository root:

make setup
make run examples/gcn.py

Other good starting points are located in examples/ (e.g. hgnn.py, villain.py).

What happens when you run an example¶

Most examples follow the same high-level pipeline:

1. Load a dataset (e.g. AlgebraDataset).
2. Split into train/val/test.
3. Add negative samples.
4. Enrich node features (e.g. Laplacian positional encoding).
5. Create one or more models (HLP modules).
6. Train + evaluate with MultiModelTrainer.

Artifacts are written under hyperbench_logs/ (by default). In particular:

• hyperbench_logs/experiment_*/comparison/results.md (markdown table).
• hyperbench_logs/experiment_*/comparison/results.tex (LaTeX table).

Next steps¶

• Dataset ingestion: HIF integration.
• Model selection/customization: Models.
• Training loop (callbacks, devices, etc.): Training.
• Comparing multiple models consistently: Benchmarking.
• Outputs and logging: Loggers.
• Visualizing runs: TensorBoard.