This paper introduces torch-uncertainty, a unified PyTorch-based framework that benchmarks state-of-the-art uncertainty quantification methods across multiple deep learning tasks and modalities.