{
  "train_micro_batch_size_per_gpu": 1,
  "gradient_accumulation_steps": 1,

  "bf16": { "enabled": true },
  "fp16": { "enabled": false },

  "zero_optimization": {
    "stage": 3,
    "overlap_comm": true,
    "contiguous_gradients": true,
    "reduce_bucket_size": 500000000,
    "stage3_prefetch_bucket_size": 200000000,
    "stage3_param_persistence_threshold": 1000000,
    "offload_optimizer": { "device": "none" },
    "offload_param":     { "device": "none" }
  },

  "gradient_clipping": 1.0,
  "wall_clock_breakdown": false
}