scripts/ds_config_zero.json

{
    "train_micro_batch_size_per_gpu": 4,
    "gradient_accumulation_steps": 1,
    "steps_per_print": 1,
    "gradient_clipping": 0.1,
    "zero_optimization": {
      "stage": 2,
      "cpu_offload": true,
      "contiguous_gradients": false,
      "overlap_comm": true,
      "reduce_scatter": false,
      "reduce_bucket_size": 100000000,
      "allgather_bucket_size": 1000000000,
      "load_from_fp32_weights": false
    },
    "zero_allow_untested_optimizer": true,
    "fp16": {
      "enabled": true,
      "loss_scale": 0,
      "loss_scale_window": 400,
      "hysteresis": 2,
      "min_loss_scale": 1
    },
    "optimizer": {
      "type": "Adam",
      "params": {
        "lr": 0.0002,
        "betas": [
          0.9,
          0.95
        ],
        "eps": 1e-8,
        "weight_decay": 1e-4
      }
    },
    "activation_checkpointing": {
      "partition_activations": false,
      "contiguous_memory_optimization": false
    },
    "wall_clock_breakdown": false
  }