Tag
TorchJD is a library for training models with multiple losses in PyTorch, implementing both scalarization and Jacobian descent methods. It has been accepted into the PyTorch ecosystem and aims to become the go-to library for multi-loss training.