Optimizer

You can define your custom optimizer by pointing the optimizer._target_ key to a class that extends from torch.optim.Optimizer. You can configure the optimizers by specifying their constructor arguments under the optimizer config key. Take a look at config/optimizer/adamw.yaml for an example.

To use custom optimizer classes, simply extend from torch.optim.Optimizer and use said class in your configuration value. You can either define your optimizer inside a custom yaml file under the configs/optimizer and include it by specifying the new optimizer via the newly created filename:

optimizer: <your-new-optimizer.yaml> 

You can also simply extend your experiment configuration file and add the optimizer._target key that points to the fully qualified Python package name of your optimizer.