質問編集履歴
1
コード,マシンスペックの追加
test
CHANGED
File without changes
|
test
CHANGED
@@ -16,6 +16,80 @@
|
|
16
16
|
|
17
17
|
|
18
18
|
|
19
|
+
### コード (pseudo)
|
20
|
+
|
21
|
+
コードの概要部分のみをpseudo code likeに記します.
|
22
|
+
|
23
|
+
|
24
|
+
|
25
|
+
```python 3.7
|
26
|
+
|
27
|
+
import torch
|
28
|
+
|
29
|
+
from torch.utils.data import DataLoader
|
30
|
+
|
31
|
+
from torch.optim import Adam
|
32
|
+
|
33
|
+
from torch.cuda import amp
|
34
|
+
|
35
|
+
import gc
|
36
|
+
|
37
|
+
|
38
|
+
|
39
|
+
train_set, valid_set, test_set = load_dataset( input_file )
|
40
|
+
|
41
|
+
|
42
|
+
|
43
|
+
train_loader = DataLoader( dataset=train_set, batch_size=batch_size )
|
44
|
+
|
45
|
+
valid_loader = DataLoader( dataset=valid_set, batch_size=batch_size )
|
46
|
+
|
47
|
+
test_loader = DataLoader( dataset=test_set, batch_size=batch_size )
|
48
|
+
|
49
|
+
|
50
|
+
|
51
|
+
mode = MyModel( parameters )
|
52
|
+
|
53
|
+
|
54
|
+
|
55
|
+
loss_criterion = torch.nn.BCEWithLogitsLoss( reduction='none' )
|
56
|
+
|
57
|
+
oprimizer = Adam( model.parameters(), lr=..., weight_decay=... )
|
58
|
+
|
59
|
+
stopper = EarlyStopper( parameters )
|
60
|
+
|
61
|
+
|
62
|
+
|
63
|
+
model.to( 'cuda:0' )
|
64
|
+
|
65
|
+
|
66
|
+
|
67
|
+
for epoch in range( n_epochs ):
|
68
|
+
|
69
|
+
|
70
|
+
|
71
|
+
train_score = train( model, train_loader, loss_criterion, optimizer )
|
72
|
+
|
73
|
+
valid_score = eval( model, valid_loader )
|
74
|
+
|
75
|
+
|
76
|
+
|
77
|
+
early_stop = stopper.step( valid_score, model )
|
78
|
+
|
79
|
+
gc.collect()
|
80
|
+
|
81
|
+
|
82
|
+
|
83
|
+
print( train_score, valid_score )
|
84
|
+
|
85
|
+
|
86
|
+
|
87
|
+
```
|
88
|
+
|
89
|
+
学習自体は問題なく進むのですが,epochが進むにつれてじわじわとRAM使用率が上昇していってしまいます.
|
90
|
+
|
91
|
+
|
92
|
+
|
19
93
|
### 環境など
|
20
94
|
|
21
95
|
|
@@ -25,3 +99,11 @@
|
|
25
99
|
PyTorch 1.7.0
|
26
100
|
|
27
101
|
DGL 0.6
|
102
|
+
|
103
|
+
|
104
|
+
|
105
|
+
CPU Intel Core i9
|
106
|
+
|
107
|
+
RAM 32GB
|
108
|
+
|
109
|
+
GPU NVidia GeForce RTX 3070
|