adam_fp16.json

{
  "train_batch_size": 16,
  "gradient_accumulation_steps": 1,
  "optimizer": {
    "type": "Adam",
    "params": {
      "torch_adam": true,
      "lr": 0.00004
    }
  },
  "fp16": {
    "enabled": true,
    "auto_cast": true
  },
  "steps_per_print": 50
}