current args description in `instruction_following.py` is not updated to our current training script.

[Luodian]

We may need to update them in our next PR.

e.g.

    parser.add_argument("--use_media_placement_augmentation", action="store_true")
    parser.add_argument("--offline", action="store_true")
    parser.add_argument("--num_epochs", type=int, default=1)
    parser.add_argument("--logging_steps", type=int, default=100, help="log loss every n steps")
    # Sum of gradient optimization batch size
    parser.add_argument("--batch_size", type=int, default=128)

    parser.add_argument("--gradient_accumulation_steps", type=int, default=1)
    parser.add_argument(
        "--pretrained_model_name_or_path",
        type=str,
        help="path to huggingface model or model identifier from local path or huggingface.co",
        default=None,
    )
    parser.add_argument(
        "--load_from_original_checkpoint",
        type=str,
        help="path to openflamingo provided checkpoint, in .pt format",
        default=None,
    )
    parser.add_argument(
        "--resume_from_checkpoint",
        action="store_true",
    )
    parser.add_argument(
        "--overwrite_checkpoint",
        action="store_true",
    )
    parser.add_argument(
        "--delete_previous_checkpoint",
        action="store_true",
        help="delete previous checkpoint when saving new checkpoint",
    )
    parser.add_argument(
        "--multi_instruct_path",
        type=str,
        help="path to multi_instruct dataset, this should be a glob pattern such as vision_language_examples.tsv",
    )
    parser.add_argument(
        "--images_path",
        type=str,
        help="path to images_path dataset, this should be a glob pattern such as vision_language_examples.tsv",
    )
    parser.add_argument(
        "--train_config_path",
        type=str,
        help="path to train_config_path dataset, this should be a glob pattern such as vision_language_examples.tsv",
    )
    parser.add_argument("--seed", type=int, default=42)
    parser.add_argument("--learning_rate", default=1e-4, type=float)
    parser.add_argument(
        "--lr_scheduler",
        default="constant",
        type=str,
        help="constant, linear, or cosine",
    )
    parser.add_argument("--loss_multiplier_multi_instruct", type=float, default=1.0)

[Luodian]

159