best_params.json

{
  "learning_rate": 7.788857568884925e-06,
  "weight_decay": 0.05084547421893973,
  "warmup_ratio": 0.0018410240823107159,
  "batch_size": 8,
  "grad_accum": 8
}