Gradient-Compression/argparser.py at master · D-J-Harris/Gradient-Compression · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
import argparse

# separate file for defining arguments from terminal
def get_args():

    parser = argparse.ArgumentParser(description='LSTM model, Gradient Compression')

    parser.add_argument('--data', type=str, default='./penn',
                        help='location of the data corpus')

    parser.add_argument('--hidden_size', type=int, default=300,
                        help='size of word embeddings/hidden size LSTM')

    parser.add_argument('--num_workers', type=int, default=1,
                        help='number of workers simulated')

    parser.add_argument('--seq_length', type=int, default=35,
                        help='backpropagation through time parameter')

    parser.add_argument('--num_layers', type=int, default=2,
                        help='number of LSTM layers (>1 for dropout)')

    parser.add_argument('--batch_size_train', type=int, default=256,
                        help='batch size during training')

    parser.add_argument('--batch_size_test', type=int, default=1,
                        help='batch size during testing')

    parser.add_argument('--num_epochs', type=int, default=50,
                        help='number of epochs')

    parser.add_argument('--dropout_prob', type=float, default=0.0,
                        help='dropout probability, regularisation')

    parser.add_argument('--tie_weights', type=bool, default=True,
                        help='tie weights of in_ and out_embeddings')

    parser.add_argument('--initial_lr', type=float, default=20.0,
                        help='initial learning rate')

    parser.add_argument('--cuda', type=bool, default=True,
                        help='default use CUDA')

    parser.add_argument('--project_name', type=str, default="project_name",
                        help='project name for wandb instance')

    parser.add_argument('--seed', type=int, default=42,
                        help='random seed for consistent testing')

    parser.add_argument('--compression', type=str, default='none',
                        help='method used for gradient compression')

    parser.add_argument('--memory', type=str, default='none',
                        help='method used for memory on gradient residuals')

    parser.add_argument('--wandb', type=bool, default=True,
                        help='default use wandb for metric logging')

    parser.add_argument('--compression_ratio', type=float, default=0.01,
                        help='default compression ratio for sparsification techniques')

    parser.add_argument('--quantum_num', type=int, default=64,
                        help='number of quantisation levels used in QSGD')

    parser.add_argument('--experiment_name', type=str, default='experiment_name',
                        help='experiment name for particular wandb run')

    parser.add_argument('--patience', type=int, default=10,
                        help='patience parameter for early stopping')

    parser.add_argument('--save_model', type=str, default='./models',
                        help='path for saving model checkpointing')

    parser.add_argument('--dgc_momentum', type=float, default=0.1,
                        help='momentum applied to dgc memory correction')


    args = parser.parse_args()
    return args