| 4 | |
| 5 | |
| 6 | def get_argument_parser(): |
| 7 | parser = argparse.ArgumentParser() |
| 8 | |
| 9 | # Required_parameter |
| 10 | parser.add_argument("--config-file", "--cf", |
| 11 | help="pointer to the configuration file of the experiment", type=str, required=True) |
| 12 | parser.add_argument("--output_dir", default=None, type=str, required=True, |
| 13 | help="The output directory where the model checkpoints will be written.") |
| 14 | |
| 15 | # Optional Params |
| 16 | parser.add_argument("--max_seq_length", default=512, type=int, |
| 17 | help="The maximum total input sequence length after WordPiece tokenization. Sequences " |
| 18 | "longer than this will be truncated, and sequences shorter than this will be padded.") |
| 19 | parser.add_argument("--max_predictions_per_seq", "--max_pred", default=80, type=int, |
| 20 | help="The maximum number of masked tokens in a sequence to be predicted.") |
| 21 | parser.add_argument('--seed', |
| 22 | type=int, |
| 23 | default=42, |
| 24 | help="random seed for initialization") |
| 25 | |
| 26 | parser.add_argument("--do_lower_case", |
| 27 | default=True, |
| 28 | action='store_true', |
| 29 | help="Whether to lower case the input text. True for uncased models, False for cased models.") |
| 30 | parser.add_argument("--local_rank", |
| 31 | type=int, |
| 32 | default=-1, |
| 33 | help="local_rank for distributed training on gpus") |
| 34 | |
| 35 | parser.add_argument('--use_pretrain', |
| 36 | default=False, |
| 37 | action='store_true', |
| 38 | help="Whether to use Bert Pretrain Weights or not") |
| 39 | |
| 40 | parser.add_argument('--refresh_bucket_size', |
| 41 | type=int, |
| 42 | default=1, |
| 43 | help="This param makes sure that a certain task is repeated for this time steps to \ |
| 44 | optimise on the back propogation speed with APEX's DistributedDataParallel") |
| 45 | parser.add_argument('--finetune', |
| 46 | default=False, |
| 47 | action='store_true', |
| 48 | help="Whether to finetune only") |
| 49 | |
| 50 | parser.add_argument('--load_training_checkpoint', '--load_cp', |
| 51 | type=str, |
| 52 | default=None, |
| 53 | help="This is the path to the TAR file which contains model+opt state_dict() checkpointed.") |
| 54 | parser.add_argument('--load_checkpoint_id', '--load_cp_id', |
| 55 | type=str, |
| 56 | default=None, |
| 57 | help='Checkpoint identifier to load from checkpoint path') |
| 58 | parser.add_argument('--job_name', |
| 59 | type=str, |
| 60 | default=None, |
| 61 | help="This is the path to store the output and TensorBoard results.") |
| 62 | |
| 63 | parser.add_argument('--rewarmup', |