-
Notifications
You must be signed in to change notification settings - Fork 9
/
run.py
155 lines (140 loc) · 5.5 KB
/
run.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
import argparse
from argparse import ArgumentParser
import os
import json
import random
from evaluation import evaluate
import numpy as np
import torch
import pytorch_lightning as pl
from pytorch_lightning.loggers import WandbLogger
from models import load_model
from Datasets import Pretrain
from pytorch_lightning.callbacks.model_checkpoint import ModelCheckpoint
def set_seed(seed):
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
if torch.cuda.is_available():
torch.cuda.manual_seed_all(seed)
if __name__ == '__main__':
parser = ArgumentParser()
parser.add_argument('--config', default=None, type=str)
arg_ = parser.parse_args()
if arg_.config == None:
raise NameError("Include a config file in the argument please.")
#Getting configurations
with open(arg_.config) as config_file:
hparam = json.load(config_file)
hparam = argparse.Namespace(**hparam)
#Setting GPUs to use
os.environ["CUDA_DEVICE_ORDER"]="PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"]=hparam.CUDA_VISIBLE_DEVICES
#Logging into WANDB if needed
if hparam.wandb_log:
wandb_logger = WandbLogger(project=hparam.wandb_project, name=hparam.wandb_run_name, entity="lklab_kaist")
else:
wandb_logger = None
#Init configs that are not given
if 'split_num' not in hparam:
hparam.split_num = 1
if 'split' not in hparam:
hparam.split = 0
if 'grad_norm' not in hparam:
hparam.grad_norm = 0.5
if 'weight_decay' not in hparam:
hparam.weight_decay = 0.0
if 'output_log' not in hparam:
hparam.output_log = None
#Setting configurations
args_dict = dict(
output_dir=hparam.output_dir, # Path to save the checkpoints
dataset=hparam.dataset,
dataset_version = hparam.dataset_version,
split_num = hparam.split_num,
split = hparam.split,
model_name_or_path=hparam.model,
method=hparam.method,
freeze_level=hparam.freeze_level,
mode=hparam.mode,
tokenizer_name_or_path=hparam.model,
max_input_length=hparam.input_length,
max_output_length=hparam.output_length,
freeze_encoder=False,
freeze_embeds=False,
learning_rate=hparam.learning_rate,
weight_decay=hparam.weight_decay,
adam_epsilon=1e-8,
warmup_steps=0,
train_batch_size=hparam.train_batch_size,
eval_batch_size=hparam.train_batch_size,
num_train_epochs=hparam.num_train_epochs,
gradient_accumulation_steps=hparam.gradient_accumulation_steps,
n_gpu=hparam.ngpu,
num_workers=hparam.num_workers,
resume_from_checkpoint=hparam.resume_from_checkpoint,
use_lr_scheduling = hparam.use_lr_scheduling,
val_check_interval = 1.0,
n_val=-1,
n_train=-1,
n_test=-1,
early_stop_callback=False,
use_deepspeed=hparam.use_deepspeed,
opt_level='O1', # you can find out more on optimisation levels here https://nvidia.github.io/apex/amp.html#opt-levels-and-properties
max_grad_norm=hparam.grad_norm, # if you enable 16-bit training then set this to a sensible value, 0.5 is a good default
seed=42,
check_validation_only=hparam.check_validation,
checkpoint_path=hparam.checkpoint_path,
accelerator=hparam.accelerator,
output_log=hparam.output_log,
)
args = argparse.Namespace(**args_dict)
# Defining how to save model checkpoints during training. Details: https://pytorch-lightning.readthedocs.io/en/stable/api/pytorch_lightning.callbacks.model_checkpoint.html
callbacks = [ModelCheckpoint(dirpath = args.output_dir, save_top_k=-1, period=1)]
checkpoint_callback = True
if args.output_dir=="":
checkpoint_callback = False # Do not save model checkpoints when output dir is empty
callbacks=[]
# Logging Learning Rate Scheduling
if args.use_lr_scheduling and hparam.wandb_log:
callbacks.append(pl.callbacks.LearningRateMonitor())
if args.use_deepspeed:
plugins = 'deepspeed_stage_2'
use_fp_16 = True
else:
plugins = []
use_fp_16 = False
# Setting Flags for pytorch lightning trainer. Details: https://pytorch-lightning.readthedocs.io/en/stable/common/trainer.html#trainer-flags
train_params = dict(
accumulate_grad_batches=args.gradient_accumulation_steps,
plugins=plugins,
gpus=args.n_gpu,
max_epochs=args.num_train_epochs,
precision= 16 if use_fp_16 else 32,
amp_level=args.opt_level,
resume_from_checkpoint=args.resume_from_checkpoint,
gradient_clip_val=args.max_grad_norm,
checkpoint_callback=checkpoint_callback,
val_check_interval=args.val_check_interval,
logger=wandb_logger,
callbacks = callbacks,
accelerator=args.accelerator,
)
#Getting the Model type & Method
if 't5' in args.model_name_or_path:
model_type='T5'
elif 'gpt2' in args.model_name_or_path:
model_type='GPT2'
else:
raise Exception('Select the correct model. Supporting "t5" and "gpt2" only.')
Model = load_model(type=model_type)
if args.check_validation_only:
evaluate(args, Model)
else:
set_seed(40)
if args.checkpoint_path!="":
model = Model.load_from_checkpoint(checkpoint_path=args.checkpoint_path, hparams=args, strict=False)
else:
model = Model(args)
trainer = pl.Trainer(**train_params)
trainer.fit(model)