-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
165 lines (127 loc) · 5.58 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
import os
import re
import random
import torch
import transformers
import numpy as np
import pandas as pd
from tqdm import tqdm
from torch.utils.data import Dataset, DataLoader
from sklearn.model_selection import train_test_split
from datasets import load_dataset
from transformers import AutoModel, AutoTokenizer, DataCollatorWithPadding
from model.t5_model import CumulativeT5Model
import wandb
import emoji
import evaluate
from soynlp.normalizer import repeat_normalize
from nltk.tokenize import sent_tokenize
os.environ["CUDA_VISIBLE_DEVICES"] = "1"
device = 'cuda' if torch.cuda.is_available() else 'cpu'
CFG = {
'model_name' :"google/flan-t5-small",
'dataset_name' :"cnn_dailymail",
'train_epoch' :20,
'lr' :1e-5,
'batch_size' :16,
'max_length' :128,
'seed' :2023,
}
def seed_everything(seed):
random.seed(seed)
os.environ['PYTHONHASHSEED'] = str(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = True
seed_everything(CFG['SEED']) # Seed 고정
# 1. Load dataset
dataset = load_dataset(CFG['dataset_name'], '3.0.0')
tokenizer = AutoTokenizer.from_pretrained(CFG['model_name'], use_faset=False)
# dataset['train'] = dataset['train'].select(list(range(0,64000)))
# dataset['validation'] = dataset['validation'].select(list(range(0,6400)))
# 2. Preprocess dataset
def preprocess_func(example):
x_key = ['article', 'highlights'] # rename to dataset column
pattern = re.compile(f'[^ .,?!/@$%~%·∼()\x00-\x7Fㄱ-ㅣ가-힣\[\]]+')
url_pattern = re.compile(r'https?:\/\/(www\.)?[-a-zA-Z0-9@:%._\+~#=]{1,256}\.[a-zA-Z0-9()]{1,6}\b([-a-zA-Z0-9()@:%_\+.~#?&//=]*)')
for key in x_key:
x = example[key]
x = pattern.sub(' ', x)
x = emoji.replace_emoji(x, replace='') #emoji 삭제
x = url_pattern.sub('', x)
x = x.strip()
x = repeat_normalize(x, num_repeats=2)
example[key] = x
del example['id'] # column 삭제
return example
def tokenizer_func(example):
model_inputs = tokenizer(example['article'], truncation=True, max_length=CFG['max_length'])
labels = tokenizer(text_target=example['highlights'], truncation=True, max_length=CFG['max_length'])
model_inputs['labels'] = labels['input_ids']
return model_inputs
cleaned_dataset = dataset.map(preprocess_func, num_proc=96)
tokenized_dataset = cleaned_dataset.map(tokenizer_func, num_proc=96, batched=True)
tokenized_dataset = tokenized_dataset.remove_columns(['article', 'highlights'])
tokenized_dataset.set_format("torch")
print('tokenized_dataset:', tokenized_dataset)
# 3. Load model and dataset
model = CumulativeT5Model.from_pretrained(CFG['model_name']).to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=CFG['lr'])
from transformers import DataCollatorForSeq2Seq
data_collator =DataCollatorForSeq2Seq(tokenizer=tokenizer, model=model, label_pad_token_id=0)
train_dataloader = DataLoader(tokenized_dataset['train'], batch_size=CFG['batch_size'], shuffle=False, collate_fn=data_collator)
valid_dataloader = DataLoader(tokenized_dataset['validation'], batch_size=CFG['batch_size'], shuffle=False, collate_fn=data_collator)
wandb.init(project='t5_sum',tags=['py', CFG['model_name']],)
wandb.watch(model)
# # 4. Train model
def train(model, dataloader):
train_loss = []
model.train()
for data in tqdm(dataloader):
data = data.to(device)
output = model(input_ids=data['input_ids'], labels=data['labels'])
loss = output.loss
loss.backward()
optimizer.step()
optimizer.zero_grad()
train_loss.append(loss.item())
# wandb.log({'loss_tracking':train_loss})
avg_loss = np.mean(train_loss)
return avg_loss
# 5. Validatoin model
def postprocess_text(preds, labels):
preds = [pred.strip() for pred in preds]
labels = [label.strip() for label in labels]
# rougeLSum expects newline after each sentence
preds = ["\n".join(sent_tokenize(pred)) for pred in preds]
labels = ["\n".join(sent_tokenize(label)) for label in labels]
return preds, labels
def validation(model, dataloader):
metric = evaluate.load("rouge")
valid_loss, avg_rouge_score = [], []
model.eval()
for data in tqdm(dataloader):
data = data.to(device)
preds = model(input_ids=data['input_ids'], labels=data['labels'])
loss = preds.loss
valid_loss.append(loss.item())
decoded_preds = tokenizer.batch_decode(torch.argmax(preds.logits, dim=2), skip_special_tokens=True)
decoded_labels = tokenizer.batch_decode(data['labels'], skip_special_tokens=True)
decoded_preds, decoded_labels = postprocess_text(decoded_preds, decoded_labels)
result = metric.compute(predictions=decoded_preds, references=decoded_labels, use_stemmer=True)
result = {k: round(v * 100, 4) for k, v in result.items()}
prediction_lens = [np.count_nonzero(pred != tokenizer.pad_token_id) for pred in preds]
result["gen_len"] = np.mean(prediction_lens)
avg_rouge_score.append(result)
avg_rouge = pd.DataFrame(avg_rouge_score).mean().to_dict()
avg_loss = np.mean(valid_loss)
return avg_loss, avg_rouge
# 6. Run
for epoch in range(CFG['train_epoch']):
train_loss = train(model, train_dataloader)
valid_loss, valid_rouge = validation(model, valid_dataloader)
print(f'Epoch:{epoch}, Train_loss:{train_loss}, Valid_loss:{valid_loss}')
wandb.log({'train_loss':train_loss, 'valid_loss':valid_loss})
wandb.log(valid_rouge)