/
sweep.py
175 lines (152 loc) · 6.95 KB
/
sweep.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
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
This script can be used to search the best hyper parameters for training.
"""
import os
import logging
import wandb
from argparse import ArgumentParser
from cli import parser, process_args
from utils import set_seed
from model import TransformerModelWrapper
from config import load_pet_configs
from data_utils import TRAIN_SET, DEV_SET, DEV32_SET, TEST_SET, load_examples, load_metrics
logger = logging.getLogger('sweep')
def main():
# Initialize wandb
run = wandb.init(reinit=True, sync_tensorboard=True)
config = wandb.config
task, seed, encoder_type = config['task'], config['seed_split'], config['encoder_type']
lr, wd, bs = config['learning_rate'], config['weight_decay'], config['batch_size']
########################################
# Prepare full arguments
data_split = '16-%d' % seed
if task == 'MNLI-mm':
data_dir = os.path.join('data', 'k-shot', 'MNLI', data_split)
elif task == 'RTE-glue':
data_dir = os.path.join('data', 'k-shot', 'RTE', data_split)
else:
data_dir = os.path.join('data', 'k-shot', task, data_split)
task_dir = os.path.join('output', task, 'tune', encoder_type)
output_dir = os.path.join(task_dir, data_split)
arguments = ['--model_type', 'roberta',
'--embed_size', '1024',
'--do_train', '--do_eval',
'--eval_set', 'test',
'--overwrite_output_dir',
'--task_name', task.lower(),
'--data_dir', data_dir,
'--pet_max_steps', '250',
'--model_name_or_path', 'roberta-large',
'--cache_dir', 'pretrain/roberta-large',
'--pet_per_gpu_eval_batch_size', '8',
'--output_dir', output_dir,
'--learning_rate', str(lr),
'--weight_decay', str(wd),
'--prompt_encoder_type', encoder_type]
if task in ['MNLI', 'MNLI-mm', 'SNLI', 'RTE-glue']:
arguments.extend(['--pet_max_seq_length', '256',
'--pet_per_gpu_train_batch_size', str(bs),
'--pet_gradient_accumulation_steps', '2'])
else:
arguments.extend(['--pet_max_seq_length', '128',
'--pet_per_gpu_train_batch_size', str(bs),
'--pet_gradient_accumulation_steps', '1'])
args = parser.parse_args(arguments)
process_args(args)
logger.info(args)
########################################
# Load dataset
train_data = load_examples(args.task_name, args.data_dir, TRAIN_SET,
num_examples=args.train_examples, split_examples_evenly=args.split_examples_evenly)
eval_data = load_examples(args.task_name, args.data_dir, TEST_SET if args.eval_set == 'test' else DEV_SET,
num_examples=args.eval_examples, split_examples_evenly=args.split_examples_evenly)
dev_data = load_examples(args.task_name, args.data_dir, DEV32_SET,
num_examples=args.dev_examples, split_examples_evenly=args.split_examples_evenly)
########################################
# Training process
set_seed(args.seed)
# Load model
model_config, train_config, eval_config = load_pet_configs(args)
model = TransformerModelWrapper(model_config)
# Train model
model.train(train_data=train_data,
dev_data=dev_data,
eval_data=eval_data,
pattern_iter_output_dir=args.output_dir,
eval_config=eval_config,
per_gpu_train_batch_size=train_config.per_gpu_train_batch_size,
n_gpu=train_config.n_gpu,
num_train_epochs=train_config.num_train_epochs,
max_steps=args.pet_max_steps,
gradient_accumulation_steps=train_config.gradient_accumulation_steps,
weight_decay=args.weight_decay,
learning_rate=args.learning_rate,
fix_other_embeddings=False,
wandb_log=False)
run.finish()
if __name__ == '__main__':
run_parser = ArgumentParser()
run_parser.add_argument("--task",
type=str,
choices=['SST-2', 'sst-5', 'mr', 'cr', 'mpqa', 'subj', 'trec', 'CoLA',
'MNLI', 'MNLI-mm', 'SNLI', 'QNLI', 'RTE-glue', 'MRPC', 'QQP'])
run_parser.add_argument("--encoder",
type=str,
default='inner',
choices=['none', 'mlp', 'lstm', 'inner', 'inner2'])
run_parser.add_argument("--seed_split",
type=int,
default=[],
nargs='+',
choices=[13, 21, 42, 87, 100])
run_parser.add_argument("--batch_size",
type=int,
default=[],
nargs='+',
choices=[4, 8, 16, 24, 32])
run_parser.add_argument("--sweep_id",
type=str,
default='')
run_args = run_parser.parse_args()
if not run_args.seed_split: # Default search all seed splits
run_args.seed_split = [13, 21, 42, 87, 100]
if not run_args.batch_size: # Default search all batch sizes
if run_args.task in ['MNLI', 'MNLI-mm', 'SNLI', 'RTE-glue']:
# Restrict maximum batch size due to memory limit
run_args.batch_size = [4, 8, 16]
else:
run_args.batch_size = [4, 8, 16, 24, 32]
# Prepare sweep config and get sweep id
sweep_config = {
'program': run_args.task,
'method': 'grid',
'metric': {
'goal': 'maximize',
'name': 'eval-' + load_metrics(run_args.task)[-1]
},
'parameters': {
'task': {'value': run_args.task},
'encoder_type': {'value': run_args.encoder},
'seed_split': {'values': run_args.seed_split},
'learning_rate': {'values': [1e-5, 5e-5, 1e-4, 2e-4]},
'weight_decay': {'values': [0.0, 0.01, 0.05, 0.10]},
'batch_size': {'values': run_args.batch_size}
}
}
if run_args.sweep_id: # Recover from old sweep
sweep_id = run_args.sweep_id
else: # Create new sweep
sweep_id = wandb.sweep(sweep_config, project="L-tune")
# Sweep all hyper parameters
wandb.agent(sweep_id, function=main)