-
Notifications
You must be signed in to change notification settings - Fork 0
/
pretrain_wan.py
304 lines (272 loc) · 15.3 KB
/
pretrain_wan.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
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
import os
import sys
sys.path.append('/root/autodl-tmp/') # 如果你是在autodl上训练，否则改成你的环境父目
import time
import random
from dataclasses import dataclass, field
import numpy as np
import pandas as pd
import torch
from transformers import (
DataCollatorForLanguageModeling,
Trainer,
TrainerCallback,
TrainingArguments,
)
import argparse
from transformers.trainer_callback import TrainerControl, TrainerState
from datasets import Dataset, load_dataset
from model.wan_config import WANConfig
config = WANConfig()
from model.modeling_wan import WANForCausalLM
from transformers import AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained('/root/autodl-tmp/tokenize_me', trust_remote_code=True)
os.environ["TF_ENABLE_ONEDNN_OPTS"] = "0"
tokenizer.pad_token_id = tokenizer.eos_token_id
sky_train = ['/root/autodl-tmp/data/sky_new/2020-50_zh_middle_0008.parquet',
'/root/autodl-tmp/data/sky_new/2020-50_zh_middle_0006.parquet',
'/root/autodl-tmp/data/sky_new/2020-50_zh_middle_0000.parquet',
'/root/autodl-tmp/data/sky_new/2020-50_zh_middle_0002.parquet',
'/root/autodl-tmp/data/sky_new/2020-50_zh_head_0009.parquet',
'/root/autodl-tmp/data/sky_new/2020-50_zh_head_0008.parquet',
'/root/autodl-tmp/data/sky_new/2020-50_zh_head_0007.parquet',
'/root/autodl-tmp/data/sky_new/2020-50_zh_head_0005.parquet',
'/root/autodl-tmp/data/sky_new/2020-50_zh_head_0004.parquet',
'/root/autodl-tmp/data/sky_new/2020-50_zh_head_0003.parquet',
'/root/autodl-tmp/data/sky_new/2020-50_zh_head_0002.parquet',
'/root/autodl-tmp/data/sky_new/2020-50_zh_head_0001.parquet',
'/root/autodl-tmp/data/sky_new/2020-50_zh_head_0000.parquet',
'/root/autodl-tmp/data/sky_new/2020-50_zh_head_0006.parquet',
'/root/autodl-tmp/data/sky_new/2020-45_zh_middle_0013.parquet',
'/root/autodl-tmp/data/sky_new/2020-45_zh_middle_0007.parquet',
'/root/autodl-tmp/data/sky_new/2020-45_zh_middle_0004.parquet',
'/root/autodl-tmp/data/sky_new/2020-45_zh_middle_0002.parquet',
'/root/autodl-tmp/data/sky_new/2020-45_zh_middle_0001.parquet',
'/root/autodl-tmp/data/sky_new/2020-45_zh_middle_0010.parquet',
'/root/autodl-tmp/data/sky_new/2020-45_zh_middle_0011.parquet',
'/root/autodl-tmp/data/sky_new/2020-45_zh_middle_0009.parquet',
'/root/autodl-tmp/data/sky_new/2020-45_zh_middle_0008.parquet',
'/root/autodl-tmp/data/sky_new/2020-45_zh_middle_0006.parquet',
'/root/autodl-tmp/data/sky_new/2020-45_zh_middle_0003.parquet',
'/root/autodl-tmp/data/sky_new/2020-45_zh_middle_0000.parquet',
'/root/autodl-tmp/data/sky_new/2020-45_zh_middle_0005.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_head_0001.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_head_0002.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_head_0009.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_head_0004.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_head_0011.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_head_0003.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_head_0005.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_head_0006.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_head_0007.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_head_0010.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_head_0014.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_head_0013.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_head_0016.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_middle_0000.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_middle_0003.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_middle_0002.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_middle_0005.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_middle_0006.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_middle_0009.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_middle_0010.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_middle_0008.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_middle_0011.parquet',
'/root/autodl-tmp/data/sky_new/2020-40_zh_head_0015.parquet']
code_train = ['/root/autodl-tmp/data/code_new/train-00004-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00005-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00007-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00008-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00019-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00018-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00021-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00010-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00009-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00012-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00011-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00013-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00014-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00015-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00016-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00017-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00020-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00022-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00028-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00029-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00023-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00026-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00027-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00025-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00030-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00024-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00031-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00032-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00033-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00034-of-00880.parquet',
'/root/autodl-tmp/data/code_new/train-00035-of-00880.parquet']
wiki_en_train = ['/root/autodl-tmp/data/wiki_en_new/train-00007-of-00042-3877aae3fec33422.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00009-of-00042-c9908579eeaf4175.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00013-of-00042-af0983718d50d5b0.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00006-of-00042-9efd9286dc070409.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00014-of-00042-02ca0950368ff729.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00008-of-00042-b851398463f73678.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00010-of-00042-81271737f1362069.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00005-of-00042-180bf29cdace3aa1.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00002-of-00042-a930e3e6733eeedd.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00004-of-00042-bbae804ae18a9fd4.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00003-of-00042-a17e9f8763b78cdf.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00001-of-00042-7aefbfce07009caf.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00000-of-00042-fb075d4d7bcac9ca.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00015-of-00042-4c861e51c53c1271.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00016-of-00042-53643c66a84d0263.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00017-of-00042-b0ce5a338667aec0.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00020-of-00042-44a4db60e9884736.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00022-of-00042-19e84dddc15d3df9.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00018-of-00042-4299b93584494629.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00019-of-00042-e6e41f11f581c2f9.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00021-of-00042-92e38583c5e1ed11.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00024-of-00042-1ba9e84168529381.parquet',
'/root/autodl-tmp/data/wiki_en_new/train-00023-of-00042-9a33c5735dcf5db7.parquet']
random.seed(42) # 设置随机种子
# 根据您的数据集情况自己设置
TRAIN_FILES = sky_train[len(sky_train) // 2:] + code_train[len(code_train) // 2:] + wiki_en_train[
len(wiki_en_train) // 2:]
random.shuffle(TRAIN_FILES) # 打乱顺序
vocab_size = len(tokenizer)
if vocab_size % 64 != 0:
vocab_size = (vocab_size // 64 + 1) * 64
print(f"final vocab sieze: {vocab_size}")
# ## token to id缓存到文件，使用的时候不用再次tokenize
# 如果词表大小小于 65535 用uint16存储，节省磁盘空间，否则用uint32存储
map_dtype = np.uint16 if vocab_size < 65535 else np.uint32
def token_to_id(samples: dict) -> dict:
batch_txt = samples["text"]
outputs = tokenizer(
batch_txt,
truncation=False,
padding=False,
return_attention_mask=False,
)
input_ids = [np.array(item, dtype=map_dtype) for item in outputs["input_ids"]]
return {"input_ids": input_ids}
print(token_to_id({'text': ['判断给定的文章是否符合语法规则。如果不符合，请提供修改建议。\n',
'下面是一篇文章的开头: "为了探讨这个主题，本文将提供一系列数据和实例，以证明这一观点。']}))
def get_maped_dataset(files) -> Dataset:
dataset = load_dataset(
path="parquet",
data_files=files,
split="train",
cache_dir=".cache",
keep_in_memory=False,
)
print("Loaded dataset size:", len(dataset))
# 确保 dataset 不为空
if len(dataset) == 0:
raise ValueError("Dataset is empty. Please check the data files and their content.")
maped_dataset = dataset.map(
token_to_id,
batched=True,
batch_size=10000,
remove_columns=dataset.column_names,
num_proc=45,
keep_in_memory=False,
)
return maped_dataset
train_dataset = get_maped_dataset(TRAIN_FILES)
print(train_dataset)
data_collator = DataCollatorForLanguageModeling(tokenizer, mlm=False)
config = WANConfig()
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = WANForCausalLM(config).to(
device
)
model_size_shared_experts = sum(t.numel() for name, t in model.named_parameters() if 'shared' in name)
model_size_act_experts = sum(
t.numel() for name, t in model.named_parameters() if 'shared' not in name and 'experts' in name)
# print(f"当前MOE模型共享专家层的激活参数大约为：{ model_size_shared_experts / 10000**2/10:.1f}B parameters")
# print(f"当前MOE模型激活专家层的激活参数大约为：{ model_size_act_experts / 10000**2/10/(config.n_routed_experts//config.num_experts_per_tok):.1f}B parameters")
# print(f"当前MOE模型的总参数大小为: {model_size / 10000**2/10:.1f}B parameters")
model_size = sum(t.numel() for t in model.parameters())
print(f"当前模型总参数大小为: {model_size / 10000 ** 2 / 10:.1f}B ")
# print(f"当前MOE模型的总参数大小为: {model_size / 10000**2/10:.1f}B parameters")
# # 6. cuda cache回调函数
class MyTrainerCallback(TrainerCallback):
log_cnt = 0
def on_log(
self,
args: TrainingArguments,
state: TrainerState,
control: TrainerControl,
**kwargs,
):
"""
在打印 n 次日志后清除cuda缓存，适合低显存设备，能防止OOM
"""
self.log_cnt += 1
if self.log_cnt % 2 == 0:
torch.cuda.empty_cache()
def on_epoch_end(
self,
args: TrainingArguments,
state: TrainerState,
control: TrainerControl,
**kwargs,
):
"""
在on_epoch_end时保存一次模型。
TrainingArguments的 save_strategy 中 epoch 和 steps 不兼容。要实现每隔 save_steps 步保存一次检查点，考虑到磁盘空间大小，最多只保存最近3个检查点。
"""
# 设置should_save=True并返回即可
control.should_save = True
return control
def collate_fn(batch):
batch = [item.to(device) for item in batch]
return torch.utils.data.dataloader.default_collate(batch)
my_trainer_callback = MyTrainerCallback()
def main():
parser = argparse.ArgumentParser(description="Run training")
parser.add_argument("--model_save_dir", type=str, default="./model_save")
parser.add_argument("--train_batch_size", type=int, default=20)
parser.add_argument("--gradient_accumulation_steps", type=int, default=10)
parser.add_argument("--num_train_epochs", type=int, default=1)
parser.add_argument("--weight_decay", type=float, default=0.1)
parser.add_argument("--learning_rate", type=float, default=1e-4)
parser.add_argument("--save_steps", type=int, default=50)
parser.add_argument("--logging_steps", type=int, default=20)
parser.add_argument("--warmup_steps", type=int, default=1000)
args = parser.parse_args()
training_args = TrainingArguments(
output_dir=args.model_save_dir,
per_device_train_batch_size=args.train_batch_size,
gradient_accumulation_steps=args.gradient_accumulation_steps,
num_train_epochs=args.num_train_epochs,
weight_decay=args.weight_decay,
ddp_find_unused_parameters=False,
warmup_steps=args.warmup_steps,
learning_rate=args.learning_rate,
save_steps=args.save_steps,
save_strategy="steps",
save_total_limit=3,
report_to="tensorboard",
optim="adamw_torch",
lr_scheduler_type="cosine",
bf16=True,
logging_steps=args.logging_steps,
log_level="info",
logging_first_step=True,
)
trainer = Trainer(
model=model,
tokenizer=tokenizer,
args=training_args,
data_collator=data_collator,
train_dataset=train_dataset,
callbacks=[my_trainer_callback],
)
# trainer.train(resume_from_checkpoint='./model_save/pre/checkpoint-3762') #如果是断点训练
trainer.train()
loss_log = pd.DataFrame(trainer.state.log_history)
loss_log.to_csv(f"./logs/pre_train_log_{time.strftime('%Y%m%d-%H%M')}.csv")
trainer.save_model(args.model_save_dir)
if __name__ == "__main__":
main()