optim:
  _target_: torch.optim.SGD
  lr: 1e-3
  weight_decay: 0.01

exclude_ln_and_biases_from_weight_decay: False