model_training_args.json

{
  "model_type": "gpt2",
  "model_checkpoint": "gpt2",
  "device": "cuda:0",
  "use_amp": true,
  "train_batch_size": 3,
  "valid_batch_size": 3,
  "gradient_accumulation_steps": 10,
  "lr": 6.25e-05,
  "max_norm": 1.0,
  "n_epochs": 3,
  "eval_before_start": false,
  "dataset_cache_dir": "/home/ubuntu/question-generation/caches/datasets",
  "dataset_dir": "datasets/SQuAD/qg",
  "local_rank": 0,
  "output_dir": "/home/ubuntu/question-generation/outputs/train",
  "save_interval_epoch": 1,
  "iterations_before_checking": 0,
  "n_saved_epoch": 3,
  "distributed": true
}