CHOP is a library for continuous and constrained optimization based on PyTorch with applications to training neural networks and adversarial examples. It includes both stochastic optimizers and full-gradient optimizers that operate on a batch of optimization problems (and can be applied for adversarial attacks).