optimizer_3x.yml 274 B

1234567891011121314151617
  1. epoch: 36
  2. LearningRate:
  3. base_lr: 0.000025
  4. schedulers:
  5. - !PiecewiseDecay
  6. gamma: 0.1
  7. milestones: [28, 34]
  8. - !LinearWarmup
  9. start_factor: 0.01
  10. steps: 1000
  11. OptimizerBuilder:
  12. clip_grad_by_norm: 1.0
  13. optimizer:
  14. type: AdamW
  15. weight_decay: 0.0001