adam_zero2_bf16.json

{
  "train_batch_size": 16,
  "gradient_accumulation_steps": 1,
  "optimizer": {
    "type": "Adam",
    "params": {
      "lr": 0.00004
    }
  },
  "bf16": {
    "enabled": true
  },
  "zero_optimization": {
    "stage": 1,
    "reduce_bucket_size": 5e8,
    "overlap_comm": true
  },
  "log": {
    "steps_per_print": 50
  }
}