model:
  transformer_model: "microsoft/deberta-v3-large"

optimizer:
  lr: 
    - 0.0001
    - 0.00002
  warmup_steps: 500
  total_steps: ${training.trainer.max_steps}
  total_reset: 1
  weight_decay: 0.01
  lr_decay: 0.9
  no_decay_params:
    - "bias"
    - LayerNorm.weight    
  other_lr_params:
    - "re_subject_projector"
    - "re_object_projector"
    - "re_relation_projector"
    - "re_classifier"
relations_per_forward: 24
entities_per_forward: