Harness

NeuroBench harness

The NeuroBench harness is an open-source Python package for standardized benchmark execution across neuromorphic tasks.

• GitHub: NeuroBench/neurobench
• Documentation: Read the Docs
• PyPI: neurobench

Evaluation pipeline

A benchmark run follows a modular flow:

Data -> Preprocess -> Model -> Postprocess -> Metrics -> Results

This structure keeps benchmark definitions readable and enables consistent reporting.

Core components

• neurobench.benchmarks: benchmark orchestration
• neurobench.datasets: benchmark datasets
• neurobench.models: wrappers for Torch and SNNTorch models
• neurobench.preprocessing: preprocessing interfaces
• neurobench.postprocessing: postprocessing interfaces
• neurobench.metrics: static and workload metrics

Quick start

pip install neurobench

Then follow the API and tutorial docs:

• API overview
• Tutorial index
• Contributing guide

Citation

Use the project citation from the paper page:

• NeuroBench paper (Nature Communications)
• Project DOI