| 33 | |
| 34 | |
| 35 | def parse_option(): |
| 36 | parser = argparse.ArgumentParser( |
| 37 | 'InternVL training and evaluation script', add_help=False) |
| 38 | parser.add_argument('--cfg', type=str, required=True, metavar='FILE', help='path to config file') |
| 39 | parser.add_argument('--opts', help="Modify config options by adding 'KEY VALUE' pairs. ", default=None, nargs='+') |
| 40 | |
| 41 | # easy config modification |
| 42 | parser.add_argument('--batch-size', type=int, help='batch size for single GPU') |
| 43 | parser.add_argument('--dataset', type=str, help='dataset name', default=None) |
| 44 | parser.add_argument('--data-path', type=str, help='path to dataset') |
| 45 | parser.add_argument('--zip', action='store_true', help='use zipped dataset instead of folder dataset') |
| 46 | parser.add_argument('--cache-mode', type=str, default='part', choices=['no', 'full', 'part'], |
| 47 | help='no: no cache, ' |
| 48 | 'full: cache all data, ' |
| 49 | 'part: sharding the dataset into nonoverlapping pieces and only cache one piece' |
| 50 | ) |
| 51 | parser.add_argument('--pretrained', help='pretrained weight from checkpoint, could be imagenet22k pretrained weight') |
| 52 | parser.add_argument('--resume', help='resume from checkpoint') |
| 53 | parser.add_argument('--output', default='work_dirs', type=str, metavar='PATH', |
| 54 | help='root of output folder, the full path is <output>/<model_name>/<tag> (default: output)' |
| 55 | ) |
| 56 | parser.add_argument('--eval', action='store_true', help='Perform evaluation only') |
| 57 | parser.add_argument('--throughput', action='store_true', help='Test throughput only') |
| 58 | parser.add_argument('--save-ckpt-num', default=1, type=int) |
| 59 | parser.add_argument('--accumulation-steps', type=int, default=1, help='gradient accumulation steps') |
| 60 | parser.add_argument('--disable-grad-scalar', action='store_true', help='disable Grad Scalar') |
| 61 | parser.add_argument( |
| 62 | '--logger', |
| 63 | type=str, |
| 64 | default='tensorboard', |
| 65 | choices=['tensorboard', 'wandb'], |
| 66 | help=( |
| 67 | 'Whether to use [tensorboard](https://www.tensorflow.org/tensorboard) or [wandb](https://www.wandb.ai)' |
| 68 | ' for experiment tracking and logging of model metrics and model checkpoints' |
| 69 | ), |
| 70 | ) |
| 71 | |
| 72 | args, unparsed = parser.parse_known_args() |
| 73 | config = get_config(args) |
| 74 | config.defrost() |
| 75 | config.TRAIN.OPTIMIZER.USE_ZERO = False |
| 76 | config.OUTPUT += '_deepspeed' |
| 77 | config.DATA.IMG_ON_MEMORY = False |
| 78 | config.freeze() |
| 79 | return args, config |
| 80 | |
| 81 | |
| 82 | def seed_everything(seed, rank): |