conf/ds_stage2.json

{
  "train_micro_batch_size_per_gpu": 1,
  "gradient_accumulation_steps": 8,
  "steps_per_print": 10,
  "gradient_clipping": 5,
  "fp16": {
    "enabled": false,
    "auto_cast": true,
    "loss_scale": 0,
    "initial_scale_power": 16,
    "loss_scale_window": 1000,
    "hysteresis": 2,
    "consecutive_hysteresis": false,
    "min_loss_scale": 1
  },
  "bf16": {
    "enabled": true
  },
  "zero_force_ds_cpu_optimizer": false,
  "zero_optimization": {
    "stage": 2,
    "offload_optimizer": {
      "device": "none",
      "pin_memory": true
    },
    "allgather_partitions": true,
    "allgather_bucket_size": 2e8,
    "reduce_scatter": true,
    "reduce_bucket_size": 2e8,
    "contiguous_gradients": false,
    "overlap_comm": false
  },
  "find_unused_parameters": true
}