ds_config_decaylr.json

{
  "bf16": {
    "enabled": "auto"
  },
  "optimizer": {
    "type": "AdamW",
    "params": {
      "lr": "auto",
      "betas": "auto",
      "eps": "auto",
      "weight_decay": "auto"
    }
  },
  "scheduler": {
      "type": "WarmupDecayLR",
      "params": {
          "last_batch_iteration": 1,
          "total_num_steps": "auto",
          "warmup_max_lr":"auto",
          "warmup_min_lr":2e-5,
          "warmup_num_steps": "auto"
      }
  }, 
  "zero_optimization": {
    "stage": 2,
    "contiguous_gradients": true,
    "sub_group_size": 1e9,
    "allgather_partitions": true,
    "allgather_bucket_size": 2e8,
    "overlap_comm": true,
    "reduce_scatter": true,
    "reduce_bucket_size": 2e8
  },
  "activation_checkpointing": {
    "partition_activations": true,
    "cpu_checkpointing": false,
    "contiguous_memory_optimization": false,
    "synchronize_checkpoint_boundary": false,
    "profile": true
  },
  "gradient_accumulation_steps": "auto",
  "gradient_clipping": "auto",
  "steps_per_print": 2000,
  "train_batch_size": "auto",
  "train_micro_batch_size_per_gpu": "auto",
  "wall_clock_breakdown": false
}