😎 Bensemble

A modular, lightweight PyTorch library for Bayesian Deep Learning and Neural Network Ensembling.

  • PyTorch-Native

    No hidden black-box wrappers. Train models using standard PyTorch loops or Lightning, and plug Bensemble in for inference and analytics.

  • Uncertainty Analytics

    Principled decomposition of predictive uncertainty into Aleatoric and Epistemic components.

  • Model Calibration

    Built-in post-hoc calibration tools like Temperature Scaling and Vector Scaling to fix overconfident neural networks.

  • Ensemble Search

    Algorithms to automatically search for diverse architectures using NNI and Stein Variational Gradient Descent.