forked from boostcampaitech2/klue-level2-nlp-15
-
Notifications
You must be signed in to change notification settings - Fork 0
/
config.yaml
113 lines (101 loc) · 2.44 KB
/
config.yaml
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
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
data:
root_path: ./
root_data_path: ./data
train_file_path: ./data/train.csv
test_file_path: ./data/test_data.csv
prediction_data_path: ./result
label_to_num_file_path: ./data/dict_label_to_num.pkl
num_to_label_file_path: ./data/dict_num_to_label.pkl
pororo_train_path: ./data/train_pororo_sub.csv
pororo_test_path: ./data/test_pororo_sub.csv
pororo_special_token_path: ./data/pororo_special_token.txt
result_dir: ./result
saved_model_dir: ./best_models
logging_dir: ./logs
IB:
# wandb configuration
user_name: happyface
# dataset configuration
num_labels: 30
num_workers: 4
max_token_length: 132
stopwords: []
# train configuration
pretrained_model_name: klue/roberta-large
fine_tuning_method: ib
batch_size: 16
num_folds: 5
num_train_epochs: 5
loss: focalloss
gamma: 1
optimizer: adamp
scheduler: get_linear_schedule_with_warmup
learning_rate: 0.00005
weight_decay: 0.01
gradient_accumulation_steps: 2
max_grad_norm: 1
warmup_ratio: 0.1
warmup_steps: 500
debug: false
dropout_rate: 0.1
# evaluation and saving configuration
save_steps: 100
evaluation_steps: 100
metric_for_best_model: f1
load_best_model_at_end: true
RBERT:
# wandb configuration
user_name: happyface
# dataset configuration
num_labels: 30
num_workers: 4
max_token_length: 132
stopwords: []
# train configuration
pretrained_model_name: klue/roberta-large
fine_tuning_method: RBERT
batch_size: 40
num_folds: 5
num_train_epochs: 5
loss: focalloss
gamma: 1
optimizer: adamp
scheduler: get_cosine_schedule_with_warmup
learning_rate: 0.00005
weight_decay: 0.01
warmup_steps: 500
debug: false
dropout_rate: 0.1
# evaluation and saving configuration
save_steps: 100
evaluation_steps: 100
metric_for_best_model: eval/loss
load_best_model_at_end: true
Concat:
# wandb configuration
user_name: happyface
# dataset configuration
num_labels: 30
num_workers: 4
max_token_length: 132
stopwords: []
# train configuration
pretrained_model_name: klue/roberta-large
fine_tuning_method: concat
batch_size: 40
num_folds: 5
num_train_epochs: 3
loss: focalloss
gamma: 0.5
optimizer: adamp
learning_rate: 0.00005
weight_decay: 0.01
warmup_steps: 300
debug: false
dropout_rate: 0.1
# evaluation and saving configuration
save_steps: 100
evaluation_strategy: steps
evaluation_steps: 500
metric_for_best_model: accuracy
load_best_model_at_end: true