scripts/zero2.json

{
    "fp16": {
        "enabled": "auto",
        "loss_scale": 0,
        "loss_scale_window": 1000,
        "initial_scale_power": 16,
        "hysteresis": 2,
        "min_loss_scale": 1
    },
    "bf16": {
        "enabled": "auto"
    },
    "train_micro_batch_size_per_gpu": "auto",
    "train_batch_size": "auto",
    "gradient_accumulation_steps": "auto",
    "zero_optimization": {
        "stage": 2,
        "overlap_comm": true,
        "contiguous_gradients": true,
        "sub_group_size": 1e8,
        "reduce_bucket_size": 2e8
    }
}