-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathresume_train_dynamic_mask.py
191 lines (149 loc) · 9.16 KB
/
resume_train_dynamic_mask.py
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
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
# @Author: Khush Patel, Zhi lab
#The idea is to change the seed of mask (run_no_mask) every few epochs. This is used for resuming training from last checkpoint saved.
# local imports
import model as m
from config import *
from dataset_MLM import *
from engine_amp_MLM import *
if __name__ == "__main__":
writer = SummaryWriter(tbpath)
with open(raw_data_path, "rb") as f:
rawdata = pickle.load(f)
mbdataset = MBDataset(rawdata, seed_value_changer=run_no_mask)
loader = torch.utils.data.DataLoader(
mbdataset, batch_size=batch_size, shuffle=True, num_workers= 12, pin_memory=True)
model = m.model
model = model.to(device)
optim = AdamW(model.parameters(), lr=lr)
scheduler = get_linear_schedule_with_warmup(
optim, num_warmup_steps=0, num_training_steps=num_train_steps
)
checkpoint = torch.load(save_dir, pickle_module=dill)
model.load_state_dict(checkpoint['model_state_dict'])
optim.load_state_dict(checkpoint['optimizer_state_dict'])
last_epoch = checkpoint['epoch']
loss = checkpoint['loss']
scheduler = checkpoint['scheduler']
steps_completed = checkpoint['number_of_training_steps']
remainingepochs = num_of_epochs - last_epoch
print(remainingepochs)
best_loss = np.inf
#Commenting out already completed epochs
# for epoch in range(num_of_epochs):
# train_loss= train_fn(loader, model, optimizer=optim, device=device, scheduler=scheduler, epoch=epoch, writer=writer,seed_value_changer=run_no_mask, steps_completed=steps_completed)
# print(f"Training loss at epoch {epoch} is {train_loss}")
# if train_loss<best_loss:
# best_loss = train_loss
# torch.save(model.state_dict(), save_wts_loc)
# print(f"Lowest training loss found at epoch {epoch}. Saving the model weights")
# ####Changing mask seed
# print("Changing the mask")
# mbdataset = MBDataset(rawdata, seed_value_changer=run_no_mask+1)
# loader = torch.utils.data.DataLoader(
# mbdataset, batch_size=batch_size, shuffle=True, num_workers= 12, pin_memory=True)
# for epoch in range(num_of_epochs):
# train_loss = train_fn(loader, model, optimizer=optim, device=device, scheduler=scheduler, epoch=epoch, writer=writer, seed_value_changer=run_no_mask+1, steps_completed=steps_completed)
# print(f"Training loss at epoch {epoch} is {train_loss}")
# if train_loss<best_loss:
# best_loss = train_loss
# torch.save(model.state_dict(), save_wts_loc)
# print(f"Lowest training loss found at epoch {epoch}. Saving the model weights")
# ####Changing mask seed
# print("Changing the mask")
# mbdataset = MBDataset(rawdata, seed_value_changer=run_no_mask+2)
# loader = torch.utils.data.DataLoader(
# mbdataset, batch_size=batch_size, shuffle=True, num_workers= 12, pin_memory=True)
# for epoch in range(num_of_epochs):
# train_loss = train_fn(loader, model, optimizer=optim, device=device, scheduler=scheduler, epoch=epoch, writer=writer, seed_value_changer=run_no_mask+2, steps_completed=steps_completed)
# print(f"Training loss at epoch {epoch} is {train_loss}")
# if train_loss<best_loss:
# best_loss = train_loss
# torch.save(model.state_dict(), save_wts_loc)
# print(f"Lowest training loss found at epoch {epoch}. Saving the model weights")
####Changing mask seed
print("Changing the mask")
mbdataset = MBDataset(rawdata, seed_value_changer=run_no_mask+3)
loader = torch.utils.data.DataLoader(
mbdataset, batch_size=batch_size, shuffle=True, num_workers= 12, pin_memory=True)
for epoch in range(last_epoch+1, num_of_epochs):
train_loss = train_fn(loader, model, optimizer=optim, device=device, scheduler=scheduler, epoch=epoch, writer=writer, seed_value_changer=run_no_mask+3, steps_completed=steps_completed)
print(f"Training loss at epoch {epoch} is {train_loss}")
if train_loss<best_loss:
best_loss = train_loss
torch.save(model.state_dict(), save_wts_loc)
print(f"Lowest training loss found at epoch {epoch}. Saving the model weights")
####Changing mask seed
print("Changing the mask")
mbdataset = MBDataset(rawdata, seed_value_changer=run_no_mask+4)
loader = torch.utils.data.DataLoader(
mbdataset, batch_size=batch_size, shuffle=True, num_workers= 12, pin_memory=True)
for epoch in range(num_of_epochs):
train_loss = train_fn(loader, model, optimizer=optim, device=device, scheduler=scheduler, epoch=epoch, writer=writer, seed_value_changer=run_no_mask+4, steps_completed=steps_completed)
print(f"Training loss at epoch {epoch} is {train_loss}")
if train_loss<best_loss:
best_loss = train_loss
torch.save(model.state_dict(), save_wts_loc)
print(f"Lowest training loss found at epoch {epoch}. Saving the model weights")
####Changing mask seed
print("Changing the mask")
mbdataset = MBDataset(rawdata, seed_value_changer=run_no_mask+5)
loader = torch.utils.data.DataLoader(
mbdataset, batch_size=batch_size, shuffle=True, num_workers= 12, pin_memory=True)
for epoch in range(num_of_epochs):
train_loss = train_fn(loader, model, optimizer=optim, device=device, scheduler=scheduler, epoch=epoch, writer=writer, seed_value_changer=run_no_mask+5, steps_completed=steps_completed)
print(f"Training loss at epoch {epoch} is {train_loss}")
if train_loss<best_loss:
best_loss = train_loss
torch.save(model.state_dict(), save_wts_loc)
print(f"Lowest training loss found at epoch {epoch}. Saving the model weights")
####Changing mask seed
print("Changing the mask")
mbdataset = MBDataset(rawdata, seed_value_changer=run_no_mask+6)
loader = torch.utils.data.DataLoader(
mbdataset, batch_size=batch_size, shuffle=True, num_workers= 12, pin_memory=True)
for epoch in range(num_of_epochs):
train_loss= train_fn(loader, model, optimizer=optim, device=device, scheduler=scheduler, epoch=epoch, writer=writer, seed_value_changer=run_no_mask+6, steps_completed=steps_completed)
print(f"Training loss at epoch {epoch} is {train_loss}")
if train_loss<best_loss:
best_loss = train_loss
torch.save(model.state_dict(), save_wts_loc)
print(f"Lowest training loss found at epoch {epoch}. Saving the model weights")
####Changing mask seed
print("Changing the mask")
mbdataset = MBDataset(rawdata, seed_value_changer=run_no_mask+7)
loader = torch.utils.data.DataLoader(
mbdataset, batch_size=batch_size, shuffle=True, num_workers= 12, pin_memory=True)
for epoch in range(num_of_epochs):
train_loss = train_fn(loader, model, optimizer=optim, device=device, scheduler=scheduler, epoch=epoch, writer=writer, seed_value_changer=run_no_mask+7, steps_completed=steps_completed)
print(f"Training loss at epoch {epoch} is {train_loss}")
if train_loss<best_loss:
best_loss = train_loss
torch.save(model.state_dict(), save_wts_loc)
print(f"Lowest training loss found at epoch {epoch}. Saving the model weights")
####Changing mask seed
print("Changing the mask")
mbdataset = MBDataset(rawdata, seed_value_changer=run_no_mask+8)
loader = torch.utils.data.DataLoader(
mbdataset, batch_size=batch_size, shuffle=True, num_workers= 12, pin_memory=True)
for epoch in range(num_of_epochs):
train_loss = train_fn(loader, model, optimizer=optim, device=device, scheduler=scheduler, epoch=epoch, writer=writer, seed_value_changer=run_no_mask+8, steps_completed=steps_completed)
print(f"Training loss at epoch {epoch} is {train_loss}")
if train_loss<best_loss:
best_loss = train_loss
torch.save(model.state_dict(), save_wts_loc)
print(f"Lowest training loss found at epoch {epoch}. Saving the model weights")
####Changing mask seed
print("Changing the mask")
mbdataset = MBDataset(rawdata, seed_value_changer=run_no_mask+9)
loader = torch.utils.data.DataLoader(
mbdataset, batch_size=batch_size, shuffle=True, num_workers= 12, pin_memory=True)
for epoch in range(num_of_epochs):
train_loss= train_fn(loader, model, optimizer=optim, device=device, scheduler=scheduler, epoch=epoch, writer=writer, seed_value_changer=run_no_mask+9, steps_completed=steps_completed)
print(f"Training loss at epoch {epoch} is {train_loss}")
if train_loss<best_loss:
best_loss = train_loss
torch.save(model.state_dict(), save_wts_loc)
print(f"Lowest training loss found at epoch {epoch}. Saving the model weights")
writer.flush()
writer.close()
sys.exit()