This repository has been archived by the owner on May 16, 2022. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 12
/
train.py
439 lines (380 loc) · 17.6 KB
/
train.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
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
"""
Train Transformer-VAEs using the Huggingface Trainer with Weights and Biasis.
"""
import logging
import inspect
import os
import sys
from dataclasses import dataclass, field, make_dataclass
from typing import Optional, Any
from datasets import load_dataset
import transformers
from transformers import (
AutoTokenizer,
HfArgumentParser,
TrainingArguments,
set_seed,
)
from transformers.integrations import is_wandb_available
from transformers.trainer_utils import is_main_process
from transformer_vae.trainer import VAE_Trainer
from transformer_vae.data_collator import DataCollatorForLanguageAutoencoding
from transformer_vae.trainer_callback import TellModelGlobalStep
from transformer_vae.model import Funnel_T5_VAE_Model
from transformer_vae.sequence_checks import SEQ_CHECKS
from transformer_vae.config import Funnel_T5_VAE_Config
logger = logging.getLogger(__name__)
@dataclass
class VAE_TrainingArguments(TrainingArguments):
"""
Extra arguments to specify generation during evaluation.
"""
save_steps: int = field(default=1000, metadata={"help": "Save checkpoint every X updates steps."})
generate_min_len: int = field(
default=1,
metadata={"help": "The minimum length of sequences to be generated from latent points during evaluation."},
)
generate_max_len: int = field(
default=20,
metadata={"help": "The maximum length of sequences to be generated from latent points during evaluation."},
)
seq_check: str = field(
default=None,
metadata={"help": f"Run check on sequences from random latent codes. Options: {', '.join([str(k) for k in SEQ_CHECKS.keys()])}"},
)
max_validation_size: int = field(
default=None,
metadata={"help": "Limit the eval dataset size, defaults to not limiting it, must be < validation size."},
)
sample_from_latent: bool = field(
default=False,
metadata={"help": "Whether to sample from the latent space during evaluation."},
)
test_classification: bool = field(
default=False,
metadata={"help": "Test using latent codes for unsupervised classification."},
)
cycle_loss: bool = field(
default=False,
metadata={"help": "Encourage the encoder & decoder to produce a bijective mapping. Feeds the final decoder hidden state to the encoder and compares the latent codes."},
)
vae_cycle_loss: bool = field(
default=False,
metadata={"help": "Encourage the encoder & decoder to produce a bijective mapping. Feeds the final decoder hidden state to the encoder and compares the latent codes."},
)
advisery_weight: int = field(
default=1,
metadata={"help": "Encourage the encoder & decoder to produce a bijective mapping. Feeds the final decoder hidden state to the encoder and compares the latent codes."},
)
cycle_weight: int = field(
default=1,
metadata={"help": "Encourage the encoder & decoder to produce a bijective mapping. Feeds the final decoder hidden state to the encoder and compares the latent codes."},
)
interpolate_training_step_rate: int = field(
default=1,
metadata={"help": "Run a batch of iterpolation losses every N steps."},
)
min_critic_steps: int = field(
default=1_000,
metadata={"help": "Start updating the model with the critic loss after N steps."},
)
render_text_image: bool = field(
default=False,
metadata={"help": """Render sequence as an image and log it to Weights & Biasis during interpolations.
Must be using a dataset with array_to_text & text_to_array methods.
This will override seq_check & just see if the text is a valid image."""},
)
dont_clean_up_tokenization_spaces: bool = field(
default=False,
metadata={"help": "Don't clean up token spaces, turn off for non NLP tasks."},
)
interpolate_all_at_once: bool = field(
default=False,
metadata={"help": "Treat all latent tokens as one during slerp."},
)
"""
# ModelArguments
Arguments pertaining to which model/config/tokenizer we are going to fine-tune, or train from scratch.
ModelArguments take args from Funnel_T5_VAE_Config,
"""
fields = [
(
'model_path', Optional[str], field(
default=None, metadata={"help": "The model checkpoint for weights initialization. Leave None if you want to train a model from scratch."}
)
),
(
'config_path', Optional[str], field(
default=None, metadata={"help": "Pretrained config path if not the same as model_name"}
)
),
(
'tokenizer_name', Optional[str], field(
default='t5-base', metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
)
),
(
'use_fast_tokenizer', bool, field(
default=True,
metadata={"help": "Whether to use one of the fast tokenizer (backed by the tokenizers library) or not."},
)
),
(
'cache_dir', str, field(
default=None,
metadata={"help": "Cache directory."},
)
)
] + [
(
name, type(info.default) if info.default is not None else Any, field(
default=info.default, metadata={"help": f"Has default {info.default}, see Funnel_T5_VAE_Config docstring for more info."}
)
)
# get relevent model arguments with defaults
for name, info in inspect.signature(Funnel_T5_VAE_Config.__init__).parameters.items() if name not in ['self', 'kwargs', 'use_extra_logs', 'cache_dir']
]
# ensure starting with non-default args
start_f = list(filter(lambda field: field[2].default is None, fields))
end_f = list(filter(lambda field: field[2].default is not None, fields))
ModelArguments = make_dataclass('ModelArguments', start_f + end_f)
@dataclass
class DataTrainingArguments:
"""
Arguments pertaining to what data we are going to input our model for training and eval.
"""
dataset_name: Optional[str] = field(
default=None, metadata={"help": "The name of the dataset to use (via the datasets library)."}
)
script_version: str = field(
default="main", metadata={"help": "Dataset branch to use (via the datasets library)."}
)
dataset_config_name: Optional[str] = field(
default=None, metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
)
text_column: Optional[str] = field(default=None, metadata={"help": "Use this dataset column as 'text'."})
train_file: Optional[str] = field(default=None, metadata={"help": "The input training data file (a text file)."})
validation_file: Optional[str] = field(
default=None,
metadata={"help": "An optional input evaluation data file to evaluate the perplexity on (a text file)."},
)
overwrite_cache: bool = field(default=False, metadata={"help": "Overwrite the cached training and evaluation sets"})
preprocessing_num_workers: Optional[int] = field(
default=None,
metadata={"help": "The number of processes to use for the preprocessing."},
)
mlm_probability: float = field(
default=0.0, metadata={"help": "Ratio of tokens to mask for masked language modeling loss"}
)
validation_name: str = field(
default="validation",
metadata={"help": "Name of the set to run evaluation on."},
)
classification_column: str = field(
default="class_label",
metadata={"help": "Test SVM classification using latent codes."},
)
num_classes: int = field(
default=None,
metadata={"help": "How many classes in the data, found using a ClassLabel column if none given."},
)
def __post_init__(self):
if self.dataset_name is None and self.train_file is None and self.validation_file is None:
raise ValueError("Need either a dataset name or a training/validation file.")
else:
if self.train_file is not None:
extension = self.train_file.split(".")[-1]
assert extension in ["csv", "json", "txt"], "`train_file` should be a csv, a json or a txt file."
if self.validation_file is not None:
extension = self.validation_file.split(".")[-1]
assert extension in ["csv", "json", "txt"], "`validation_file` should be a csv, a json or a txt file."
def get_args():
parser = HfArgumentParser((ModelArguments, DataTrainingArguments, VAE_TrainingArguments))
if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
# If we pass only one argument to the script and it's the path to a json file,
# let's parse it to get our arguments.
model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
else:
model_args, data_args, training_args = parser.parse_args_into_dataclasses()
# use train batch size if eval is default
training_args.per_device_eval_batch_size = (
training_args.per_device_train_batch_size
if training_args.per_device_eval_batch_size == 8
else training_args.per_device_eval_batch_size
)
if (
os.path.exists(training_args.output_dir)
and os.listdir(training_args.output_dir)
and training_args.do_train
and not training_args.overwrite_output_dir
):
raise ValueError(
f"Output directory ({training_args.output_dir}) already exists and is not empty."
"Use --overwrite_output_dir to overcome."
)
return model_args, data_args, training_args
def setup_logs(training_args):
logging.basicConfig(
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
datefmt="%m/%d/%Y %H:%M:%S",
level=logging.INFO if is_main_process(training_args.local_rank) else logging.WARN,
)
# Log on each process the small summary:
logger.warning(
f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}"
+ f", distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16}"
)
# Set the verbosity to info of the Transformers logger (on main process only):
if is_main_process(training_args.local_rank):
transformers.utils.logging.set_verbosity_info()
logger.info("Training/evaluation parameters %s", training_args)
def get_datasets(data_args):
# Get the datasets: you can either provide your own CSV/JSON/TXT training and evaluation files (see below)
# or just provide the name of one of the public datasets available on the hub at https://huggingface.co/datasets/
# (the dataset will be downloaded automatically from the datasets Hub).
#
# For CSV/JSON files, this script will use the column called 'text' or the first column if no column called
# 'text' is found. You can easily tweak this behavior (see below).
#
# In distributed training, the load_dataset function guarantee that only one local process can concurrently
# download the dataset.
if data_args.dataset_name is not None:
# Downloading and loading a dataset from the hub.
return load_dataset(data_args.dataset_name, data_args.dataset_config_name) # , script_version=data_args.script_version
data_files = {}
if data_args.train_file is not None:
data_files["train"] = data_args.train_file
if data_args.validation_file is not None:
data_files[data_args.validation_name] = data_args.validation_file
extension = data_args.train_file.split(".")[-1]
if extension == "txt":
extension = "text"
return load_dataset(extension, data_files=data_files)
# See more about loading any type of standard or custom dataset (from files, python dict, pandas DataFrame, etc) at
# https://huggingface.co/docs/datasets/loading_datasets.html.
def load_model_and_tokenizer(model_args):
# Distributed training:
# The `.from_pretrained` methods guarantee that only one local process can concurrently
# download model & vocab.
if model_args.set_seq_size and model_args.set_seq_size <= 4:
logger.warning('`set_seq_size` is to small to work with the Funnel transformer. now using set_seq_size=5')
model_args.set_seq_size = 5
if model_args.config_path:
config = Funnel_T5_VAE_Config.from_pretrained(
model_args.config_path, cache_dir=model_args.cache_dir
)
elif model_args.model_path:
config = Funnel_T5_VAE_Config.from_pretrained(
model_args.model_path, cache_dir=model_args.cache_dir
)
else:
config = Funnel_T5_VAE_Config(use_extra_logs=is_wandb_available(), **model_args.__dict__)
logger.warning("You are instantiating a new config instance from scratch (still using T5 checkpoint).")
tokenizer = AutoTokenizer.from_pretrained(
model_args.tokenizer_name, cache_dir=model_args.cache_dir, use_fast=model_args.use_fast_tokenizer
)
if model_args.model_path:
model = Funnel_T5_VAE_Model.from_pretrained(
model_args.model_path,
from_tf=bool(".ckpt" in model_args.model_path),
config=config,
cache_dir=model_args.cache_dir,
)
model.resize_token_embeddings(len(tokenizer))
else:
vocab_size = len(tokenizer)
config.funnel.vocab_size = vocab_size
config.t5.vocab_size = vocab_size
config.vocab_size = vocab_size
logger.info("Training new model from scratch")
model = Funnel_T5_VAE_Model(config)
if model_args.set_seq_size:
tokenizer.model_max_length = model_args.set_seq_size
tokenizer.mask_token = tokenizer.unk_token
return model, tokenizer
def preprocess_datasets(training_args, data_args, tokenizer, datasets):
# Add class_label if needed
if training_args.test_classification:
if data_args.classification_column != "class_label":
if not data_args.num_classes:
data_args.num_classes = (
datasets[data_args.validation_name].features[data_args.classification_column].num_classes
)
def add_class_column(examples):
return {"class_label": examples[data_args.classification_column]}
datasets = datasets.map(add_class_column, remove_columns=[data_args.classification_column])
# tokenize all the texts.
if training_args.do_train:
column_names = datasets["train"].column_names
else:
column_names = datasets[data_args.validation_name].column_names
if data_args.text_column is not None:
text_column_name = data_args.text_column
else:
text_column_name = "text" if "text" in column_names else column_names[0]
if text_column_name != "text":
logger.info(f'Using column "{text_column_name}" as text column.')
if tokenizer.pad_token_id is None:
def tokenize_function(examples):
return tokenizer(examples[text_column_name], truncation=True)
else:
def tokenize_function(examples):
return tokenizer(examples[text_column_name], padding="max_length", truncation=True)
tokenized_datasets = datasets.map(
tokenize_function,
batched=True,
batch_size=training_args.per_device_train_batch_size,
num_proc=data_args.preprocessing_num_workers,
remove_columns=[text_column_name],
load_from_cache_file=not data_args.overwrite_cache,
)
if training_args.max_validation_size:
tokenized_datasets[data_args.validation_name] = tokenized_datasets[data_args.validation_name].train_test_split(
training_args.max_validation_size
)["test"]
data_collator = DataCollatorForLanguageAutoencoding(tokenizer=tokenizer, mlm_probability=data_args.mlm_probability)
return data_collator, tokenized_datasets
def main():
model_args, data_args, training_args = get_args()
if training_args.output_dir and training_args.output_dir[-1] == '/' and training_args.run_name:
training_args.output_dir += training_args.run_name
setup_logs(training_args)
# Set seed before initializing model.
set_seed(training_args.seed)
datasets = get_datasets(data_args)
model, tokenizer = load_model_and_tokenizer(model_args)
data_collator, tokenized_datasets = preprocess_datasets(training_args, data_args, tokenizer, datasets)
# Initialize our Trainer
trainer = VAE_Trainer(
model=model,
args=training_args,
train_dataset=tokenized_datasets["train"] if training_args.do_train else None,
eval_dataset=tokenized_datasets[data_args.validation_name] if training_args.do_eval else None,
custom_methods={name: getattr(datasets, name) for name in filter(lambda x: x.startswith('custom_'), dir(datasets))},
tokenizer=tokenizer,
data_collator=data_collator,
callbacks=[TellModelGlobalStep],
)
trainer.log({})
# Training
if training_args.do_train:
trainer.train(
model_path=model_args.model_path if model_args.model_path and os.path.isdir(model_args.model_path) else None
)
trainer.save_model() # Saves the tokenizer too for easy upload
# Evaluation
results = {}
if training_args.do_eval:
logger.info("*** Evaluate ***")
results = trainer.evaluate()
output_eval_file = os.path.join(training_args.output_dir, "eval_results_T_VAE.txt")
if trainer.is_world_process_zero():
with open(output_eval_file, "w") as writer:
logger.info("***** Eval results *****")
for key, value in results.items():
logger.info(f" {key} = {value}")
writer.write(f"{key} = {value}\n")
return results
def _mp_fn(index):
# For xla_spawn (TPUs)
main()