optim:
  _target_: torch.optim.Adam
  lr: 1e-3
  betas: [0.9, 0.999]
  weight_decay: 0.01

exclude_ln_and_biases_from_weight_decay: False
lora_lr: 1e-4
backbone_lr: 5e-6
last_block_lr: 5e-5
unfreeze_lr: False
diff_backbone_last: False