ds_config.json

{
  "train_batch_size": "auto",
  "gradient_accumulation_steps": 4,
  "fp16": {
    "enabled": true
  },
  "zero_optimization": {
    "stage": 2,
    "allgather_partitions": true,
    "reduce_scatter": true,
    "allgather_bucket_size": 2e8,
    "overlap_comm": true,
    "contiguous_gradients": true
  },
  "gradient_clipping": 1.0,
  "steps_per_print": 100,
  "wall_clock_breakdown": false
}