epoch: 12

LearningRate:
  base_lr: 0.0001
  schedulers:
  - !PiecewiseDecay
    gamma: 0.1
    milestones: [9, 11]
  - !LinearWarmup
    start_factor: 0.001
    steps: 1000

OptimizerBuilder:
  optimizer:
    type: AdamWDL
    betas: [0.9, 0.999]
    layer_decay: 0.75
    weight_decay: 0.02
    num_layers: 12
    filter_bias_and_bn: True
    skip_decay_names: ['pos_embed', 'cls_token']
    set_param_lr_func: 'layerwise_lr_decay'