glue_bert_base.json

{
    "train_batch_size": 32,
    "train_micro_batch_size_per_gpu": 1,
    "steps_per_print": 10,
    "optimizer": {
      "type": "Adam",
      "params": {
        "lr": 2e-5,
        "weight_decay": 0.0,
        "bias_correction": true
      }
    },
    "gradient_clipping": 1.0,
    "fp16": {
      "enabled": false
    }
  
  }