During May 2025, this developer expanded the NVIDIA/physicsnemo library by implementing support for the CELU activation function, enhancing the range of activation options available for neural simulations. The work involved updating the ACT2FN mapping to include 'celu', linking it to the PyTorch nn.CELU class with a default alpha of 1.0, and ensuring seamless integration throughout the codebase. Using Python and deep learning frameworks, the developer added comprehensive tests to validate the new activation path. This addition improved model flexibility and reproducibility, enabling researchers to experiment with CELU as a first-class activation option within the physicsnemo repository.