-
Notifications
You must be signed in to change notification settings - Fork 0
/
vit_ft.py
275 lines (225 loc) · 7.17 KB
/
vit_ft.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
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
""" ViT fine-tuning
"""
import argparse
import functools
import sys
import traceback
import numpy as np
import torch
from evaluate import load
from torchvision.transforms import (
Compose,
Normalize,
RandomHorizontalFlip,
RandomResizedCrop,
RandomVerticalFlip,
Resize,
ToTensor,
)
from transformers import (
TrainingArguments,
ViTForImageClassification,
ViTImageProcessor,
)
import wandb
from src.config import Config
from src.dataset import NAIPImagery
from src.trainer import CustomTrainer
def load_dataset(path_to_train, path_to_test, transform=None):
"""Load data from test and train!"""
size = transform.size["height"]
# Set up transform
train_aug_transforms = Compose(
[
Resize((size, size)),
RandomResizedCrop(size=size),
RandomHorizontalFlip(p=0.5),
RandomVerticalFlip(p=0.5),
ToTensor(),
Normalize(mean=transform.image_mean, std=transform.image_std),
]
)
valid_aug_transforms = Compose(
[
Resize(size=(size, size)),
ToTensor(),
Normalize(mean=transform.image_mean, std=transform.image_std),
]
)
# Evaluation dataset
train = NAIPImagery(
images_dir=path_to_train,
transform=train_aug_transforms,
max_prompt_len=70,
tokenizer=None,
)
test = NAIPImagery(
images_dir=path_to_test,
transform=valid_aug_transforms,
max_prompt_len=70,
tokenizer=None,
)
return train, test
def model_init():
return ViTForImageClassification.from_pretrained(
"google/vit-base-patch16-224-in21k",
ignore_mismatched_sizes=True,
num_labels=2,
)
def compute_metrics(eval_pred):
# Get values from Trainer loss
logits, labels = eval_pred.predictions, eval_pred.label_ids
preds = np.argmax(logits, axis=1)
# Get metrics
metrics = dict()
accuracy_metric = load("accuracy")
precision_metric = load("precision")
recall_metric = load("recall")
f1_metric = load("f1")
metrics.update(accuracy_metric.compute(predictions=preds, references=labels))
metrics.update(
precision_metric.compute(
predictions=preds, references=labels, average="weighted"
)
)
metrics.update(
recall_metric.compute(predictions=preds, references=labels, average="weighted")
)
metrics.update(
f1_metric.compute(predictions=preds, references=labels, average="weighted")
)
return metrics
def collator(batch):
"""Stucture data for tranining"""
return {
"pixel_values": torch.cat([x["pixel_values"] for x in batch]),
"labels": torch.stack([x["labels"] for x in batch]),
}
def train(
training_dataset,
test_dataset,
image_processor,
wandb_dir,
sweep_dir,
epochs,
config=None,
):
"""Training loop for sweep"""
with wandb.init(config=config, dir=wandb_dir):
# Sweep config
config = wandb.config
try:
# Start trainer
training_args = TrainingArguments(
output_dir=sweep_dir,
learning_rate=config.learning_rate,
per_device_train_batch_size=config.batch_size,
num_train_epochs=epochs,
# gradient_accumulation_steps=4,
weight_decay=config.weight_decay,
per_device_eval_batch_size=32,
warmup_ratio=config.warmup_ratio,
logging_strategy="epoch",
evaluation_strategy="epoch",
save_strategy="epoch",
load_best_model_at_end=True,
metric_for_best_model="accuracy",
logging_dir="logs",
report_to="wandb",
fp16=False,
)
trainer = CustomTrainer(
model=model_init(),
args=training_args,
train_dataset=training_dataset,
eval_dataset=test_dataset,
tokenizer=image_processor,
compute_metrics=compute_metrics,
data_collator=collator,
)
trainer.train()
except Exception:
print(traceback.print_exc(), file=sys.stderr)
def main(config, train_fn):
"""Sweep configuration through W&B"""
config_train = config.train_config
project_name = config_train["project_name"]
model_name = config_train["model_name"]
# Add image processing
image_processor = ViTImageProcessor.from_pretrained(model_name, num_labels=2)
# Load data
training_dataset, test_dataset = load_dataset(
transform=image_processor,
path_to_train=config_train["path_to_train"],
path_to_test=config_train["path_to_test"],
)
train_fn_partial = functools.partial(
train_fn,
training_dataset,
test_dataset,
image_processor,
config_train["wandb_dir"],
config_train["sweep_dir"],
config_train["epochs"],
)
# Sweep configuration for training
sweep_configuration = {
"method": "bayes",
"name": config_train["sweep_name"],
"metric": {"goal": "minimize", "name": "eval_loss"},
"parameters": {
"batch_size": {"values": [16, 32, 64]},
"learning_rate": {"values": [5e-5, 5e-4, 2e-4]},
"weight_decay": {
"values": [0.001, 0.002, 0.005],
},
"warmup_ratio": {
"values": [0, 0.1],
},
},
}
# Set up sweep in wandb
sweep_id = wandb.sweep(sweep_configuration, project=project_name)
# Start sweep
wandb.agent(sweep_id, train_fn_partial, count=20)
def main_worker(config, train_fn, sweep_id):
"""Sweep configuration through W&B"""
config_train = config.train_config
project_name = config_train["project_name"]
model_name = config_train["model_name"]
# Add image processing
image_processor = ViTImageProcessor.from_pretrained(model_name)
# Load data
training_dataset, test_dataset = load_dataset(
transform=image_processor,
path_to_train=config_train["path_to_train"],
path_to_test=config_train["path_to_test"],
)
train_fn_partial = functools.partial(
train_fn,
training_dataset,
test_dataset,
image_processor,
config_train["wandb_dir"],
config_train["sweep_dir"],
)
# Start sweep
wandb.agent(sweep_id, train_fn_partial, count=20)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--config_file", type=str, help="Path to YAML config file")
# Init args
args = parser.parse_args()
path_to_config = args.config_file
# Init W&B
wandb.login()
############################# CUDA CONFIGURATION ##############################
device = "cpu"
if torch.cuda.device_count() > 0 and torch.cuda.is_available():
print("Cuda installed! Running on GPU!")
device = "cuda"
else:
print("No GPU available!")
###############################################################################
config = Config(path_to_config)
main(config, train)