BLiTZ is a Python library that extends PyTorch with Bayesian neural network layers implementing weight uncertainty as described in the seminal paper on Weight Uncertainty in Neural Networks. It provides variational layers where weights are sampled from learned distributions rather than fixed values, enabling uncertainty quantification in deep learning models. The library integrates seamlessly with standard PyTorch workflows, allowing users to add probabilistic modeling capabilities to existing architectures while maintaining compatibility with standard layers and training loops.