{
  "gradient_clipping": 0.0,
  "steps_per_print": 100,
  "train_batch_size": 4096,
  "train_micro_batch_size_per_gpu": 1024,
  "wall_clock_breakdown": false
}