-
Notifications
You must be signed in to change notification settings - Fork 45
/
Copy pathrun_glue_benchmark.py
172 lines (140 loc) · 7.82 KB
/
run_glue_benchmark.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
import pickle
import os
import glob
import logging
import argparse
import torch
import sys
import pandas as pd
import numpy as np
from torch.utils.data import SequentialSampler
# PROJECT_FOLDER = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
# sys.path.append(PROJECT_FOLDER)
from src import nli_data_processing
from envs import PROJECT_FOLDER, HOME_DATA_FOLDER
from BERT.pytorch_pretrained_bert.modeling import BertConfig
from BERT.pytorch_pretrained_bert.tokenization import BertTokenizer
from src.modeling import BertForSequenceClassificationEncoder, FCClassifierForSequenceClassification
from src.utils import count_parameters, load_model, eval_model_dataloader, fill_tensor
from src.data_processing import init_model, get_task_dataloader
logging.basicConfig(format='%(asctime)s - %(levelname)s - %(name)s - %(message)s',
datefmt='%m/%d/%Y %H:%M:%S',
level=logging.INFO)
logger = logging.getLogger(__name__)
DEBUG = True
ALL_TASKS = ['MRPC', 'RTE', 'SST-2', 'MNLI', 'QQP', 'MNLI-mm', 'QNLI', 'race-merge']
if DEBUG:
interested_task = 'RTE'.split(',')
prediction_mode_input = 'teacher:train,dev,test'
output_all_layers = True # True for patient teacher and False for normal teacher
bert_model = 'bert-base-uncased'
result_file = os.path.join(PROJECT_FOLDER, 'result/glue/result_summary/teacher_12layer_all.csv')
else:
interested_task = sys.argv[1].split(',')
prediction_mode_input = sys.argv[2]
output_all_layers = sys.argv[3].lower() == 'true'
bert_model = sys.argv[5]
result_file = sys.argv[4]
for t in interested_task:
assert t in ALL_TASKS, f'{t} not in all tasks! double check!'
##############################################################
# Global Variables
##############################################################
AVAILABLE_MODE = ['teacher', 'benchmark']
KD_DIR = os.path.join(HOME_DATA_FOLDER, 'outputs/KD/')
sub_dir = '_'.join(os.path.basename(result_file).split('_')[:-1])
prediction_mode, interested_set = prediction_mode_input.split(':')
assert prediction_mode in AVAILABLE_MODE, f'mode {prediction_mode} not available'
if prediction_mode == 'teacher':
output_dir = os.path.join(HOME_DATA_FOLDER, 'outputs/KD')
else:
output_dir = os.path.join(PROJECT_FOLDER, f'result/glue/benchmark/{sub_dir}')
bert_model = os.path.join(HOME_DATA_FOLDER, f'models/pretrained/{bert_model}')
config = BertConfig(os.path.join(bert_model, 'bert_config.json'))
tokenizer = BertTokenizer.from_pretrained(bert_model, do_lower_case=True)
args = argparse.Namespace(n_gpu=1,
device=torch.device("cuda" if torch.cuda.is_available() else "cpu"),
fp16=True,
eval_batch_size=32,
max_seq_length=128)
logger.info(f'sub_dir = {sub_dir}')
logger.info(f'prediction_mode = {prediction_mode}')
logger.info(f'interested_set = {interested_set}')
result_df = pd.read_csv(result_file)
for i, val in result_df.iterrows():
run_folder = val[0]
epoch = int(val[1]) - 1
information = run_folder.split('_')
task = information[1]
n_layer = int(information[2].split('.')[1])
logger.info(f'predicting for task {task}')
logger.info(f'using model from {run_folder} epoch {epoch}')
if i < 0:
logger.info('Skipped!')
continue
if interested_task is not None and task not in interested_task:
logger.info('Skipped because not interested')
continue
if 'race' in task:
args.eval_batch_size = 16
else:
args.eval_batch_size = 32
args.raw_data_dir = os.path.join(HOME_DATA_FOLDER, 'data_raw', task)
run_folder = os.path.join(KD_DIR, task, sub_dir, run_folder)
encoder_file = glob.glob(run_folder + '/*e.%d.encoder.pkl' % epoch)
cls_file = glob.glob(run_folder + '/*e.%d.cls.pkl' % epoch)
assert len(encoder_file) == 1 and len(cls_file) == 1, f'encoder/cls file error: {encoder_file}, {cls_file}'
encoder_file, cls_file = encoder_file[0], cls_file[0]
encoder_bert, classifier = init_model(task, output_all_layers, n_layer, config)
encoder_bert = load_model(encoder_bert, encoder_file, args, 'exact', verbose=True)
classifier = load_model(classifier, cls_file, args, 'exact', verbose=True)
all_res = {'train': None, 'dev': None, 'test': None}
if 'dev' in interested_set or 'valid' in interested_set:
dev_examples, dev_dataloader, dev_label_ids = get_task_dataloader(task.lower(), 'dev', tokenizer, args,
SequentialSampler, args.eval_batch_size)
dev_res = eval_model_dataloader(encoder_bert, classifier, dev_dataloader, args.device, detailed=True, verbose=False)
dev_pred_label = dev_res['pred_logit'].argmax(1)
logger.info('for dev, acc = {}, loss = {}'.format(dev_res['acc'], dev_res['loss']))
logger.info('debug dev acc = {}'.format((dev_label_ids.numpy() == dev_pred_label).mean()))
all_res['dev'] = dev_res
if 'test' in interested_set:
test_examples, test_dataloader, test_label_ids = get_task_dataloader(task.lower(), 'test', tokenizer, args,
SequentialSampler, args.eval_batch_size)
test_res = eval_model_dataloader(encoder_bert, classifier, test_dataloader, args.device, detailed=True, verbose=False)
test_pred_label = test_res['pred_logit'].argmax(1)
logger.info('for test, acc = {}, loss = {}'.format(test_res['acc'], test_res['loss']))
logger.info('debug test acc = {}'.format((test_label_ids.numpy() == test_pred_label).mean()))
if task == 'race-merge':
middle_id = np.array(['middle' in t.mrc_id for t in test_examples])
logger.info('race-middle test acc = {}'.format((test_label_ids.numpy()[middle_id] == test_pred_label[middle_id]).mean()))
logger.info('race-hight test acc = {}'.format((test_label_ids.numpy()[~middle_id] == test_pred_label[~middle_id]).mean()))
all_res['test'] = test_res
if 'train' in interested_set:
train_examples, train_dataloader, train_label_ids = get_task_dataloader(task.lower(), 'train', tokenizer, args,
SequentialSampler, args.eval_batch_size)
train_res = eval_model_dataloader(encoder_bert, classifier, train_dataloader, args.device, detailed=True, verbose=False)
train_pred_label = train_res['pred_logit'].argmax(1)
logger.info('for training, acc = {}, loss = {}'.format(train_res['acc'], train_res['loss']))
logger.info('debug train acc = {}'.format((train_label_ids.numpy() == train_pred_label).mean()))
all_res['train'] = train_res
if prediction_mode in ['benchmark']:
if 'race' in task:
continue
logger.info('saving benchmark results')
processor = nli_data_processing.processors[task.lower()]()
label_list = processor.get_labels()
test_pred_label = [label_list[tr] for tr in test_res['pred_logit'].argmax(1)]
test_pred = pd.DataFrame({'index': range(len(test_examples)), 'prediction': test_pred_label})
if task == 'MNLI':
test_pred.to_csv(os.path.join(output_dir, task + '-m.tsv'), sep='\t', index=False)
else:
test_pred.to_csv(os.path.join(output_dir, task + '.tsv'), sep='\t', index=False)
elif prediction_mode in ['teacher']:
logger.info('saving teacher results')
if not output_all_layers:
fname = os.path.join(output_dir, task, task + f'_normal_kd_teacher_{n_layer}layer_result_summary.pkl')
else:
fname = os.path.join(output_dir, task, task + f'_patient_kd_teacher_{n_layer}layer_result_summary.pkl')
with open(fname, 'wb') as fp:
pickle.dump(all_res, fp)
logger.info(f'predicting for task {task} Done!')