-
Notifications
You must be signed in to change notification settings - Fork 7
/
glue.py
255 lines (209 loc) · 8.11 KB
/
glue.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
import logging
import math
import operator
import os
from collections import defaultdict
import torch
from torch.utils.data import DataLoader, RandomSampler, TensorDataset
from transformers import (AutoConfig, AutoTokenizer, GlueDataset,
GlueDataTrainingArguments, glue_tasks_num_labels)
from autopipe.autopipe.model_profiling import (register_new_explicit_untraced_function,
register_new_traced_function)
from models.normal.NLP_models.modeling_bert import BertForSequenceClassification, get_extended_attention_mask
from models.normal.NLP_models.modeling_roberta import RobertaForSequenceClassification
from . import register_task, Parser
from .partitioning_task import PartitioningTask
logger = logging.getLogger(__name__)
MODEL_TYPES = ['bert', 'roberta']
def make_just_x(ds):
d = defaultdict(list)
for feature in ds: # no reason to go over everything...
for key, val in vars(feature).items():
if key == "label":
continue
if val is None:
continue
d[key].append(val)
print(d.keys())
return TensorDataset(*[torch.tensor(x) for x in d.values()])
# TODO: "diagnostic"
TASK_NAME_TO_DATA_DIR = {
'cola': 'CoLA',
'mnli': 'MNLI',
'mnli-mm': 'MNLI',
'mrpc': 'MPRC',
'sst-2': 'SST-2',
'sts-b': 'STS-B',
'qqp': 'QQP',
'qnli': 'QNLI',
'rte': 'RTE',
'wnli': 'WNLI'
}
def get_dataset(args, tokenizer, cache_name="glue_ds.pt"):
cache_name += args.model_name_or_path
if os.path.exists(cache_name) and not args.overwrite_cache:
print(f"-I- loading dataset from cahce {cache_name}...")
flag = False
try:
ds = torch.load(cache_name)
res = ds
except Exception as _:
print("-I- loading from cache failed. Creating new dataset. will not overwrite_cache.")
res = None
return res
print("-I- creating dataset")
data_dir = args.data_dir
task_dir = TASK_NAME_TO_DATA_DIR.get(args.task_name)
data_dir = os.path.join(data_dir, task_dir)
glue_args = GlueDataTrainingArguments(task_name=args.task_name,
data_dir=data_dir,
max_seq_length=args.max_seq_length,
overwrite_cache=args.overwrite_cache)
ds = GlueDataset(
glue_args,
tokenizer,
mode="train",
)
ds = make_just_x(ds)
if (not os.path.exists(cache_name)) or args.overwrite_cache:
print("-I- dataset saved")
torch.save(ds, cache_name)
print("-I- DONE creating dataset")
return ds
def get_sample(args, tokenizer, analysis=False):
train_dataset = get_dataset(args, tokenizer)
train_sampler = RandomSampler(train_dataset)
# TODO: create a dataloader like they do in transformers...
train_dataloader = DataLoader(train_dataset,
sampler=train_sampler,
batch_size=args.analysis_batch_size if analysis else args.partitioning_batch_size)
batch = next(iter(train_dataloader))
if args.precompute_attention_mask:
attention_mask = get_extended_attention_mask(batch[1], batch[0])
else:
attention_mask = batch[1]
inputs = {
"input_ids": batch[0],
"attention_mask": attention_mask,
}
if args.model_type == "bert":
inputs["token_type_ids"] = batch[2]
return inputs
class ParsePartitioningOptsGlue(Parser):
def _add_model_args(self, group):
group.add_argument("--task_name",
type=str,
default="mnli",
help="Glue task")
# Required parameters
group.add_argument(
"--model_type",
default=None,
type=str,
required=True,
help="Model type selected in the list: " + ", ".join(MODEL_TYPES),
)
group.add_argument(
"--model_name_or_path",
default=None,
type=str,
required=True,
help="Path to pre-trained model or shortcut name.",
)
group.add_argument(
"--precompute_attention_mask",
action="store_true",
default=False,
help="whether to compute attention mask inside or outside the model"
)
group.add_argument(
"--max_seq_length",
default=128,
type=int,
help=
"The maximum total input sequence length after WordPiece tokenization. Sequences "
"longer than this will be truncated, and sequences shorter than this will be padded.",
)
group.add_argument(
"--do_lower_case",
action="store_true",
help="Set this flag if you are using an uncased model.")
def _add_data_args(self, group):
group.add_argument(
"--data_dir",
default="/home_local/saareliad/data/glue_data/",
type=str,
help="The input data dir. Should contain the files for the task.")
group.add_argument(
"--cache_dir",
default="",
type=str,
help=
"Where do you want to store the pre-trained models downloaded from s3",
)
group.add_argument(
"--overwrite_cache",
action="store_true",
help="Overwrite the cached training and evaluation sets")
def _default_values(self):
d = {
"partitioning_batch_size": 1,
"n_iter": 1,
"n_partitions": 2,
"bw": 12,
"analysis_batch_size": 1
}
return d
def _post_parse(self, args, argv):
args.model_type = args.model_type.lower()
return super()._post_parse(args, argv)
def _auto_file_name(self, args) -> str:
bw_str = str(args.bw).replace(".", "_")
model_str = str(args.model_name_or_path).replace("-", "_")
seq_len_str = f"s_{args.max_seq_length}"
model_str += seq_len_str
output_file = f"{args.output_file}{model_str}_{args.n_partitions}p_bw{bw_str}"
if args.async_pipeline:
output_file += "_async"
output_file += f"_{args.task_name}"
output_file += "_glue"
m = args.partitioning_method.lower()
tmp = m if m != "2dbin" else "virtual_stages"
output_file += f"_{tmp}"
return output_file
class GluePartitioner(PartitioningTask):
def __init__(self, args) -> None:
super().__init__(args)
self.tokenizer = AutoTokenizer.from_pretrained(
args.model_name_or_path,
do_lower_case=args.do_lower_case,
cache_dir=args.cache_dir if args.cache_dir else None,
)
@property
def batch_dim(self) -> int:
return 0
def get_input(self, args, analysis=False):
return get_sample(args, self.tokenizer, analysis=analysis)
def get_model(self, args) -> torch.nn.Module:
config = AutoConfig.from_pretrained(args.model_name_or_path,
cache_dir=args.cache_dir if args.cache_dir else None)
setattr(config, "precompute_attention_mask", args.precompute_attention_mask)
# get correct number of labels.
config.num_labels = glue_tasks_num_labels.get(args.task_name)
model_cls = {
'bert': BertForSequenceClassification,
'roberta': RobertaForSequenceClassification
}
model_cls = model_cls[args.model_type]
model = model_cls.from_pretrained(
args.model_name_or_path,
from_tf=bool(".ckpt" in args.model_name_or_path),
config=config,
cache_dir=args.cache_dir if args.cache_dir else None,
).train()
return model
def register_functions(self):
register_new_explicit_untraced_function(operator.is_, operator)
register_new_explicit_untraced_function(operator.is_not, operator)
register_new_traced_function(math.sqrt, math)
register_task("glue", ParsePartitioningOptsGlue, GluePartitioner)