forked from LogIntelligence/LogPPT
-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
271 lines (239 loc) · 11.9 KB
/
train.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
import logging
import math
import datasets
import torch
import transformers
from datasets import load_metric
from torch.utils.data import DataLoader
from tqdm import tqdm
from transformers import (
HfArgumentParser,
AutoConfig,
default_data_collator,
get_scheduler,
set_seed
)
from logppt.models import load_model
from accelerate import Accelerator
import copy
from logppt.utils import MainArguments, ModelArguments, TrainArguments, TaskArguments, find_labels
from logppt.data import load_data_parsing, load_data_anomaly_detection, CustomDataCollator
from logppt.models import add_label_token
from logppt.tokenization import parsing_tokenize_dataset
from logppt.evaluation import evaluate
from logppt.tasks.log_parsing import template_extraction
logger = logging.getLogger(__name__)
accelerator = Accelerator()
filter_list = ["and", "or", "the", "a", "of", "to", "at"]
def train():
total_batch_size = train_args.per_device_train_batch_size * accelerator.num_processes * train_args.gradient_accumulation_steps
logger.info("***** Running training *****")
logger.info(f" Num examples = {len(train_dataset)}")
logger.info(f" Num Epochs = {train_args.num_train_epochs}")
logger.info(f" Instantaneous batch size per device = {train_args.per_device_train_batch_size}")
logger.info(f" Total train batch size (w. parallel, distributed & accumulation) = {total_batch_size}")
logger.info(f" Gradient Accumulation steps = {train_args.gradient_accumulation_steps}")
logger.info(f" Total optimization steps = {train_args.max_train_steps}")
# Only show the progress bar once on each machine.
progress_bar = tqdm(range(train_args.max_train_steps), disable=not accelerator.is_local_main_process)
completed_steps = 0
# model_path = "checkpoint_best.pt"
# early_stopping = EarlyStopping(patience=5, verbose=False, path=model_path)
for epoch in range(train_args.num_train_epochs):
model.train()
total_loss = []
for step, batch in enumerate(train_dataloader):
batch.pop('ori_labels', 'not found ner_labels')
batch = {k: v.to(device) for k, v in batch.items()}
outputs = model(**batch)
loss = outputs.loss
total_loss.append(float(loss))
loss = loss / train_args.gradient_accumulation_steps
accelerator.backward(loss)
if step % train_args.gradient_accumulation_steps == 0 or step == len(train_dataloader) - 1:
optimizer.step()
lr_scheduler.step()
optimizer.zero_grad()
progress_bar.update(1)
progress_bar.set_description(f"Loss: {float(loss)}")
completed_steps += 1
if completed_steps >= train_args.max_train_steps:
break
# early_stopping(np.average(total_loss), model)
# if early_stopping.early_stop:
# print("Early stopping!!")
# break
# load the last checkpoint with the best model
# model.load_state_dict(torch.load(model_path))
# Use the result of the last epoch
best_metric = evaluate(metric, model, tokenizer, eval_dataloader, accelerator,
main_args.pad_to_max_length,
label_token_id_to_label, id_to_label, main_args.mode)
print("Finish training, best metric: ")
try:
logger.info(best_metric)
except Exception as _:
print(best_metric)
if main_args.output_dir is not None:
accelerator.wait_for_everyone()
unwrapped_model = accelerator.unwrap_model(model)
unwrapped_model.save_pretrained(main_args.output_dir, save_function=accelerator.save)
tokenizer.save_pretrained(main_args.output_dir)
if task_args.task_name == "log-parsing":
template_extraction(tokenizer, model, accelerator, task_args.log_file, max_length=main_args.max_length,
model_name=model_type, shot=main_args.shot, dataset_name=task_args.dataset_name,
o_dir=task_args.task_output_dir, mode=main_args.mode)
else:
raise ValueError("Please choose the \"log-parsing\" task")
if __name__ == '__main__':
parser = HfArgumentParser((MainArguments, ModelArguments, TrainArguments, TaskArguments))
main_args, model_args, train_args, task_args = parser.parse_args_into_dataclasses()
logging.basicConfig(
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
datefmt="%m/%d/%Y %H:%M:%S",
level=logging.INFO,
)
logger.info(accelerator.state)
logger.setLevel(logging.INFO if accelerator.is_local_main_process else logging.ERROR)
if accelerator.is_local_main_process:
datasets.utils.logging.set_verbosity_warning()
transformers.utils.logging.set_verbosity_info()
else:
datasets.utils.logging.set_verbosity_error()
transformers.utils.logging.set_verbosity_error()
# If passed along, set the training seed now.
if main_args.seed is not None:
set_seed(main_args.seed)
# Get the datasets: the data file are JSON files
if task_args.task_name == "log-parsing":
raw_datasets, text_column_name, label_name = load_data_parsing(main_args)
else:
raw_datasets, text_column_name, label_name = load_data_anomaly_detection(main_args)
if model_args.model_name_or_path:
config = AutoConfig.from_pretrained(model_args.model_name_or_path)
model_type = config.model_type
tokenizer, model = load_model(model_args.model_name_or_path, model_type, mode=main_args.mode)
else:
raise ValueError("missing model path")
if task_args.task_name == "log-parsing":
ori_label_token_map = {"i-val": []}
else:
ori_label_token_map = {"i-normal": [], "i-abnormal": []}
sorted_add_tokens = sorted(list(ori_label_token_map.keys()), key=lambda x: len(x), reverse=True)
tokenizer.add_tokens(sorted_add_tokens)
label_list = list(ori_label_token_map.keys())
label_list += 'o'
label_to_id = {'o': 0}
for label in label_list:
if label != 'o':
label_to_id[label] = len(label_to_id)
num_labels = len(label_list)
print("ori label:", ori_label_token_map)
new_label_to_id = copy.deepcopy(label_to_id)
label_to_id = new_label_to_id
id_to_label = {id: label for label, id in label_to_id.items()}
print("label to id:", label_to_id)
print("id to label:", id_to_label)
label_token_map = {item: item for item in ori_label_token_map}
# label_token_map = ori_label_token_map
print("label token map:", label_token_map)
label_token_to_id = {label: tokenizer.convert_tokens_to_ids(label_token) for label, label_token in
label_token_map.items()}
label_token_id_to_label = {idx: label for label, idx in label_token_to_id.items()}
print("label token to id:", label_token_to_id)
print("label token id to label:", label_token_id_to_label)
# Preprocessing the datasets.
# First we tokenize all the texts.
padding = "max_length" if main_args.pad_to_max_length else False
if task_args.task_name == "log-parsing":
processed_raw_datasets, label_words, keywords = parsing_tokenize_dataset(tokenizer, raw_datasets,
text_column_name,
label_name, main_args.max_length,
padding,
label_to_id, label_token_to_id,
model_type,
main_args.mode)
train_dataset = processed_raw_datasets["train"]
eval_dataset = processed_raw_datasets["validation"]
else:
raise NotImplementedError()
print(processed_raw_datasets)
# DataLoaders creation:
if main_args.pad_to_max_length:
# If padding was already done ot max length, we use the default data collator that will just convert everything
# to tensors.
data_collator = default_data_collator
else:
# Otherwise, `CustomDataCollator` will apply dynamic padding for us (by padding to the
# maximum length of the samples passed). When using mixed precision, we add `pad_to_multiple_of=8` to pad all
# tensors to multiple of 8s, which will enable the use of Tensor Cores on NVIDIA hardware with compute
# capability >= 7.5 (Volta).
data_collator = CustomDataCollator(
tokenizer, pad_to_multiple_of=None
)
train_dataloader = DataLoader(
train_dataset, shuffle=True, collate_fn=data_collator, batch_size=train_args.per_device_train_batch_size
)
if eval_dataset is not None:
eval_dataloader = DataLoader(eval_dataset, collate_fn=data_collator,
batch_size=train_args.per_device_eval_batch_size)
else:
eval_dataloader = None
if main_args.mode == "prompt-tuning":
label_words = find_labels(model, train_dataloader, eval_dataloader, keywords)
selected_words = []
current_list = {i: 0 for i in range(100)}
lbl_word2 = copy.deepcopy(label_words)
lbl_word_indices = list(label_words.keys()).copy()
for k in lbl_word_indices:
token = tokenizer.convert_tokens_to_string([tokenizer.convert_ids_to_tokens(k)]).strip()
if k in tokenizer.all_special_ids or len(token) < 3 or token in filter_list \
or token.count(token[0]) == len(token) or token in selected_words:
del label_words[k]
else:
selected_words.append(token)
current_list[len(token.strip())] += 1
label_words = {'i-val': list(label_words.keys())[:train_args.no_label_words]}
print(tokenizer.convert_ids_to_tokens(label_words['i-val']))
model = add_label_token(model_type, model, tokenizer, label_words, train_args.wo_label_words)
# dev_dataloader = DataLoader(dev_dataset, collate_fn=data_collator, batch_size=args.per_device_eval_batch_size)
# Optimizer
# Split weights in two groups, one with weight decay and the other not.
no_decay = ["bias", "LayerNorm.weight"]
optimizer_grouped_parameters = [
{
"params": [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)],
"weight_decay": train_args.weight_decay,
},
{
"params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)],
"weight_decay": 0.0,
},
]
optimizer = torch.optim.AdamW(
optimizer_grouped_parameters,
lr=train_args.learning_rate,
betas=(0.9, 0.999),
eps=1e-8,
)
device = accelerator.device
model.to(device)
# Prepare everything with our `accelerator`.
model, optimizer, train_dataloader, eval_dataloader = accelerator.prepare(
model, optimizer, train_dataloader, eval_dataloader
)
label_id_list = torch.tensor([label_token_to_id[id_to_label[i]] for i in range(len(id_to_label)) if
i != 0 and not id_to_label[i].startswith("B-")], dtype=torch.long, device=device)
num_update_steps_per_epoch = math.ceil(len(train_dataloader) / train_args.gradient_accumulation_steps)
if train_args.max_train_steps is None:
train_args.max_train_steps = train_args.num_train_epochs * num_update_steps_per_epoch
else:
train_args.num_train_epochs = math.ceil(train_args.max_train_steps / num_update_steps_per_epoch)
lr_scheduler = get_scheduler(
name=train_args.lr_scheduler_type,
optimizer=optimizer,
num_warmup_steps=train_args.num_warmup_steps,
num_training_steps=train_args.max_train_steps,
)
metric = load_metric("logppt/evaluation/seqeval_metric.py")
train()