training_args.json

{
  "optimizer": "adamw",
  "scheduler": "onecycle",
  "learning_rate": 0.0001,
  "batch_size": 16,
  "epochs": 20,
  "seed": 123,
  "status": "default recipe; not a completed run"
}