-
Notifications
You must be signed in to change notification settings - Fork 2
/
train_generator.py
450 lines (408 loc) · 18 KB
/
train_generator.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
440
441
442
443
444
445
446
447
448
449
450
#!/usr/bin/env python
# coding=utf-8
# Copyright 2020 The HuggingFace Inc. team. All rights reserved.
#
# 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.
"""
Fine-tuning the library models for causal language modeling (GPT, GPT-2, CTRL, ...) on a text file or a dataset.
Here is the full list of checkpoints on the hub that can be fine-tuned by this script:
https://huggingface.co/models?filter=causal-lm
"""
# You can also adapt this script on your own causal language modeling task. Pointers for this are left as comments.
import logging
logging.basicConfig(level = logging.INFO)
import math
import os
import sys
from typing import Optional
from dataclasses import dataclass, field
import transformers
from src.dataset import GenerationDataset
from transformers import GlueDataTrainingArguments
from transformers import (
CONFIG_MAPPING,
MODEL_FOR_CAUSAL_LM_MAPPING,
AutoConfig,
AutoModelForCausalLM,
AutoTokenizer,
HfArgumentParser,
TrainingArguments,
default_data_collator,
set_seed,
)
from src.generation_trainer import GenTrainer as Trainer
from transformers.trainer_utils import get_last_checkpoint
from transformers.utils import check_min_version
from src.generation_model import PrefixCTRL
import torch
from src.processors import control_code_mapping, prompt_mapping, task_type_mapping, processors_mapping
# Will error if the minimal version of Transformers is not installed. Remove at your own risks.
check_min_version("4.11.0")
# require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/language-modeling/requirements.txt")
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
@dataclass
class ModelArguments:
"""
Arguments pertaining to which model/config/tokenizer we are going to fine-tune, or train from scratch.
"""
model_name_or_path: Optional[str] = field(
default=None,
metadata={
"help": "The model checkpoint for weights initialization."
"Don't set if you want to train a model from scratch."
},
)
config_overrides: Optional[str] = field(
default=None,
metadata={
"help": "Override some existing default config settings when a model is trained from scratch. Example: "
"n_embd=10,resid_pdrop=0.2,scale_attn_weights=false,summary_type=cls_index"
},
)
train_mode: str = field(
default="prefix-infix",
metadata={"help": "Generator training mode"},
)
meta_weight: bool = field(
default=False, metadata={"help": "Train model with meta weighting"}
)
prefix_len: int = field(
default=10,
metadata={"help": "Prompt length for prefix tuning"},
)
config_name: Optional[str] = field(
default=None, metadata={"help": "Pretrained config name or path if not the same as model_name"}
)
tokenizer_name: Optional[str] = field(
default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
)
cache_dir: Optional[str] = field(
default=None,
metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
)
use_fast_tokenizer: bool = field(
default=True,
metadata={"help": "Whether to use one of the fast tokenizer (backed by the tokenizers library) or not."},
)
model_revision: str = field(
default="main",
metadata={"help": "The specific model version to use (can be a branch name, tag name or commit id)."},
)
use_auth_token: bool = field(
default=False,
metadata={
"help": "Will use the token generated when running `transformers-cli login` (necessary to use this script "
"with private models)."
},
)
def __post_init__(self):
if self.config_overrides is not None and (self.config_name is not None or self.model_name_or_path is not None):
raise ValueError(
"--config_overrides can't be used in combination with --config_name or --model_name_or_path"
)
@dataclass
class DataTrainingArguments(GlueDataTrainingArguments):
"""
Arguments pertaining to what data we are going to input our model for training and eval.
"""
gen_label: str = field(default=None, metadata={"help": "The label of the generated texts."})
task_name: str = field(
default=None,
metadata={"help": "Task name"}
)
data_dir: str = field(
default=None,
metadata={"help": "Path to dataset"}
)
first_sent_limit: int = field(
default=None,
metadata={"help": "Limit the length of the first sentence (i.e., sent_0)"}
)
other_sent_limit: int = field(
default=None,
metadata={"help": "Limit the length of sentences other than the first sentence"}
)
overwrite_cache: bool = field(
default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
)
validation_split_percentage: Optional[int] = field(
default=5,
metadata={
"help": "The percentage of the train set used as validation set in case there's no validation split"
},
)
no_save: bool = field(
default=False, metadata={"help": "Do not save trained model"}
)
eval_gen_file: str = field(
default=None,
metadata={"help": "Path to generated json file to be evaluated"}
)
@dataclass
class DynamicTrainingArguments(TrainingArguments):
weight_net_lr: float = field(
default=1e-3,
metadata={"help": "learning rate of weight net"}
)
meta_lr: float = field(
default=1e-2,
metadata={"help": "learning rate of meta model"}
)
weight_net_decay: float = field(
default=1e-4,
metadata={"help": "weight decay of weight net"}
)
def main():
# See all possible arguments in src/transformers/training_args.py
# or by passing the --help flag to this script.
# We now keep distinct sets of args, for a cleaner separation of concerns.
parser = HfArgumentParser((ModelArguments, DataTrainingArguments, DynamicTrainingArguments))
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()
# Setup logging
logging.basicConfig(
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
datefmt="%m/%d/%Y %H:%M:%S",
handlers=[logging.StreamHandler(sys.stdout)],
)
log_level = training_args.get_process_log_level()
logger.setLevel(log_level)
transformers.utils.logging.set_verbosity(log_level)
transformers.utils.logging.enable_default_handler()
transformers.utils.logging.enable_explicit_format()
# 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}"
)
logger.info(f"Training/evaluation parameters {training_args}")
# Detecting last checkpoint.
last_checkpoint = None
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
last_checkpoint = get_last_checkpoint(training_args.output_dir)
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
raise ValueError(
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
"Use --overwrite_output_dir to overcome."
)
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
logger.info(
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
)
# Set seed before initializing model.
set_seed(training_args.seed)
# Load pretrained model and tokenizer
#
# Distributed training:
# The .from_pretrained methods guarantee that only one local process can concurrently
# download model & vocab.
config_kwargs = {
"cache_dir": model_args.cache_dir,
"revision": model_args.model_revision,
"use_auth_token": True if model_args.use_auth_token else None,
}
if model_args.config_name:
config = AutoConfig.from_pretrained(model_args.config_name, **config_kwargs)
elif model_args.model_name_or_path:
config = AutoConfig.from_pretrained(model_args.model_name_or_path, **config_kwargs)
tokenizer_kwargs = {
"cache_dir": model_args.cache_dir,
"use_fast": model_args.use_fast_tokenizer,
"revision": model_args.model_revision,
"use_auth_token": True if model_args.use_auth_token else None,
}
if model_args.tokenizer_name:
tokenizer = AutoTokenizer.from_pretrained(model_args.tokenizer_name, **tokenizer_kwargs)
elif model_args.model_name_or_path:
tokenizer = AutoTokenizer.from_pretrained(model_args.model_name_or_path, **tokenizer_kwargs)
else:
raise ValueError(
"You are instantiating a new tokenizer from scratch. This is not supported by this script."
"You can do it from another script, save it, and load it from here, using --tokenizer_name."
)
if tokenizer.eos_token is None:
tokenizer.add_special_tokens({'eos_token': '[EOS]'})
if tokenizer.bos_token is None:
tokenizer.add_special_tokens({'bos_token': '[BOS]'})
tokenizer.add_special_tokens({'pad_token': '[PAD]'})
task_type = task_type_mapping[data_args.task_name]
if 'infix' in model_args.train_mode and task_type == 'single':
model_args.train_mode = model_args.train_mode.replace("-infix", "")
# use the pretrained CTRL model
if model_args.model_name_or_path == 'ctrl':
label_list = list(prompt_mapping[data_args.task_name].keys())
model = PrefixCTRL.from_pretrained(
model_args.model_name_or_path,
from_tf=bool(".ckpt" in model_args.model_name_or_path),
config=config,
cache_dir=model_args.cache_dir,
revision=model_args.model_revision,
use_auth_token=True if model_args.use_auth_token else None,
task=data_args.task_name,
label_list=label_list,
prefix_len=model_args.prefix_len,
default_mode=model_args.train_mode,
bos_id=tokenizer.bos_token_id,
eos_id=tokenizer.eos_token_id,
meta_weight=model_args.meta_weight
)
# load a tuned CTRL model saved in local paths
else:
model = PrefixCTRL.from_pretrained(
model_args.model_name_or_path,
)
label_list = model.config.label_list
data_args.task_name = model.config.task
model_args.train_mode = model.config.default_mode
model.resize_token_embeddings(len(tokenizer))
control_code = control_code_mapping[data_args.task_name]
try:
task_prompt = prompt_mapping[data_args.task_name]
prefix_init = []
infix_init = []
for label in label_list:
prompt = task_prompt[label]
if type(prompt) != list:
prompt = [prompt]
if task_type == "pair" and len(prompt) == 1:
prompt = [None] + prompt
prefix_init.append(prompt[0])
if task_type == "pair":
infix_init.append(prompt[1])
else:
infix_init.append(None)
except:
prompt = None
prefix_init = None
infix_init = None
# Initialize prefix tuning parameters
if prefix_init is not None:
all_init_key_values = []
control_code_len = -1
for i, init in enumerate(prefix_init):
print('Initialization:', init)
label_control_code = control_code[label_list[i]] if type(control_code) == dict else control_code
print(f"Control code: {label_control_code}")
control_code_input = tokenizer([label_control_code])
if control_code_len == -1:
control_code_len = len(control_code_input['input_ids'][0])
else:
assert control_code_len == len(control_code_input['input_ids'][0]), "Control code length is not consistent across labels!"
init_input = tokenizer([label_control_code + ' ' + init])
output = model(input_ids=torch.tensor(init_input['input_ids']), mode="full")
print(f"Initialized prefix prompt length: {len(init_input['input_ids'][0])}")
past_key_values = output.past_key_values
init_key_values = torch.cat(past_key_values, dim=0).unsqueeze(0)
# print(init_key_values.shape)
all_init_key_values.append(init_key_values)
if training_args.do_train:
model.init_prefix_param(torch.cat(all_init_key_values, dim=0), control_code_len)
model.freeze_unoptimized_params()
control_code = None
if 'infix' in model_args.train_mode and infix_init is not None:
all_init_infix = []
for i, init in enumerate(infix_init):
init_input = tokenizer([init])
print(f"Initialized infix prompt length: {len(init_input['input_ids'][0])}")
all_init_infix.append(torch.tensor(init_input['input_ids']).unsqueeze(0))
model.init_infix_param(torch.cat(all_init_infix, dim=0))
for name, param in model.named_parameters():
if "infix" in name:
param.requires_grad = True
if "no-prompt" in model_args.train_mode or task_type == "single":
prompt = [None, None]
else:
prompt = [None, prompt[1]]
return_infix = "infix" in model_args.train_mode
train_dataset = (
GenerationDataset(data_args,
tokenizer=tokenizer,
prompt=prompt,
processor=processors_mapping[data_args.task_name],
control_code=control_code,
return_infix=return_infix,
mode="train")
)
dev_dataset = (
GenerationDataset(data_args,
tokenizer=tokenizer,
prompt=prompt,
processor=processors_mapping[data_args.task_name],
control_code=control_code,
return_infix=return_infix,
mode="dev")
if training_args.do_eval
else None
)
# Initialize our Trainer
trainer = Trainer(
model=model,
args=training_args,
train_dataset=train_dataset if training_args.do_train else None,
eval_dataset=dev_dataset if training_args.do_eval else None,
tokenizer=tokenizer,
# Data collator will default to DataCollatorWithPadding, so we change it.
data_collator=default_data_collator,
)
# Training
print(f"\n\n ### Trainable params: {[n for n, p in model.named_parameters() if p.requires_grad]} ###")
print(f"\n\n ### num of trainable params: {sum(p.numel() for p in model.parameters() if p.requires_grad)} ###")
if training_args.do_train:
checkpoint = None
if training_args.resume_from_checkpoint is not None:
checkpoint = training_args.resume_from_checkpoint
elif last_checkpoint is not None:
checkpoint = last_checkpoint
train_result = trainer.train(resume_from_checkpoint=checkpoint)
# Assign learned special token embeddings to model token embeddings
with torch.no_grad():
# Update the word embeddings for BOS and EOS
word_embeddings = trainer.model.transformer.get_input_embeddings()
word_embeddings.weight[tokenizer.bos_token_id] = trainer.model.special_emb[0]
word_embeddings.weight[tokenizer.eos_token_id] = trainer.model.special_emb[1]
trainer.model.transformer.set_input_embeddings(word_embeddings)
# Update the biases, if the model is CTRL
if 'CTRL' in type(model).__name__:
with torch.no_grad():
trainer.model.lm_head.bias[tokenizer.bos_token_id] = trainer.model.special_lm_head.bias[0]
trainer.model.lm_head.bias[tokenizer.eos_token_id] = trainer.model.special_lm_head.bias[1]
if not data_args.no_save:
trainer.save_model() # Saves the tokenizer too for easy upload
metrics = train_result.metrics
trainer.log_metrics("train", metrics)
trainer.save_metrics("train", metrics)
trainer.save_state()
# Evaluation
if training_args.do_eval:
logger.info("*** Evaluate ***")
metrics = trainer.evaluate()
try:
perplexity = math.exp(metrics["eval_loss"])
except OverflowError:
perplexity = float("inf")
metrics["perplexity"] = perplexity
trainer.log_metrics("eval", metrics)
trainer.save_metrics("eval", metrics)
kwargs = {"finetuned_from": model_args.model_name_or_path, "tasks": "text-generation"}
if training_args.push_to_hub:
trainer.push_to_hub(**kwargs)
else:
trainer.create_model_card(**kwargs)
if __name__ == "__main__":
main()