-
Notifications
You must be signed in to change notification settings - Fork 1
/
train-task.py
430 lines (348 loc) · 15.3 KB
/
train-task.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
import argparse
import json
import logging
import os
import sys
from random import Random
import datasets
import evaluate
import numpy as np
import torch
import torch.distributed as dist
import torch.multiprocessing as mp
import valohai
from accelerate import Accelerator
from datasets import load_dataset
from torch.optim import AdamW
from torch.utils.data import DataLoader
from tqdm import tqdm
from transformers import (
AutoModelForSeq2SeqLM,
AutoTokenizer,
DataCollatorForSeq2Seq,
get_scheduler,
)
import helpers
os.environ['TRANSFORMERS_NO_ADVISORY_WARNINGS'] = 'true'
class Partition:
def __init__(self, data, index):
self.data = data
self.index = index
def __len__(self):
return len(self.index)
def __getitem__(self, index):
data_idx = self.index[index]
return self.data[data_idx]
class DataPartitioner:
def __init__(self, dataset, sizes=None, seed=1234):
if sizes is None:
sizes = [0.7, 0.2, 0.1]
self.data = dataset
self.partitions = []
rng = Random()
rng.seed(seed)
data_len = len(dataset)
indexes = [x for x in range(0, data_len)]
rng.shuffle(indexes)
for frac in sizes:
part_len = int(frac * data_len)
self.partitions.append(indexes[0:part_len])
indexes = indexes[part_len:]
def use(self, partition):
return Partition(self.data, self.partitions[partition])
def average_gradients(model):
size = float(dist.get_world_size())
for param in model.parameters():
dist.all_reduce(param.grad.data, op=dist.ReduceOp.SUM)
param.grad.data /= size
class ModelTrainer:
def __init__(self, model_ckpt, batch_size=1, num_epochs=1, warmup_steps=500, evaluation_steps=500):
self.model_ckpt = model_ckpt
self.batch_size = batch_size
self.num_epochs = num_epochs
self.warmup_steps = warmup_steps
self.evaluation_steps = evaluation_steps
self.device = None
self.accelerator = Accelerator()
self.print_gpu_report()
self.tokenizer = AutoTokenizer.from_pretrained(self.model_ckpt)
self.pretrained_model = AutoModelForSeq2SeqLM.from_pretrained(self.model_ckpt)
self.logger = logging.getLogger(__name__)
self.set_logs()
def set_logs(self):
self.logger.info(self.accelerator.state)
self.logger.setLevel(logging.INFO if self.accelerator.is_local_main_process else logging.ERROR)
if self.accelerator.is_local_main_process:
datasets.utils.logging.set_verbosity_warning()
else:
datasets.utils.logging.set_verbosity_error()
def print_gpu_report(self):
from subprocess import call
print('torch.cuda.device_count() ', torch.cuda.device_count())
print('__Python VERSION:', sys.version)
print('__pyTorch VERSION:', torch.__version__)
print('__CUDA VERSION')
print('__CUDNN VERSION:', torch.backends.cudnn.version())
print('__Number CUDA Devices:', torch.cuda.device_count())
print('__Devices')
call(
[
"nvidia-smi",
"--format=csv",
"--query-gpu=index,name,driver_version,memory.total,memory.used,memory.free",
],
)
print('Active CUDA Device: GPU', torch.cuda.current_device())
print('Available devices ', torch.cuda.device_count())
print('Current cuda device ', torch.cuda.current_device())
def generate_batch_sized_chunks(self, list_of_elements):
"""split the dataset into smaller batches that we can process simultaneously
Yield successive batch-sized chunks from list_of_elements."""
for i in range(0, len(list_of_elements), self.batch_size):
yield list_of_elements[i : i + self.batch_size]
def calculate_metric_on_test_ds(self, dataset, metric):
article_batches = list(self.generate_batch_sized_chunks(dataset['article']))
target_batches = list(self.generate_batch_sized_chunks(dataset['highlights']))
for article_batch, target_batch in tqdm(zip(article_batches, target_batches), total=len(article_batches)):
inputs = self.tokenizer(
article_batch,
max_length=1024,
truncation=True,
padding="max_length",
return_tensors="pt",
)
summaries = self.pretrained_model.generate(
input_ids=inputs["input_ids"].to(self.device),
attention_mask=inputs["attention_mask"].to(self.device),
length_penalty=0.8,
num_beams=8,
max_length=128,
)
decoded_summaries = [
self.tokenizer.decode(s, skip_special_tokens=True, clean_up_tokenization_spaces=True) for s in summaries
]
decoded_summaries = [d.replace("", " ") for d in decoded_summaries]
metric.add_batch(predictions=decoded_summaries, references=target_batch)
score = metric.compute()
return score
def convert_examples_to_features(self, example_batch):
input_encodings = self.tokenizer(
example_batch['dialogue'],
padding="max_length",
truncation=True,
max_length=1024,
)
target_encodings = self.tokenizer(
text_target=example_batch['summary'],
padding="max_length",
truncation=True,
max_length=128,
)
return {
'input_ids': input_encodings['input_ids'],
'attention_mask': input_encodings['attention_mask'],
'labels': target_encodings['input_ids'],
}
def partition_dataset(self, preprocesed_dataset, collator):
import math
size = dist.get_world_size()
bsz = math.ceil((2 / float(size)))
partition_sizes = [1.0 / size for _ in range(size)]
partition = DataPartitioner(preprocesed_dataset, partition_sizes)
partition = partition.use(dist.get_rank())
set = DataLoader(
partition,
batch_size=bsz,
shuffle=True,
collate_fn=collator,
)
print('batch_size', bsz)
return set, bsz
def synchronize_and_aggregate_metrics(self, metrics):
if metrics is None:
return None
group = dist.group.WORLD
group_size = dist.get_world_size(group)
tensor_dict = {key: torch.zeros_like(torch.tensor(metrics[key])) for key in metrics.keys()}
metrics = {key: torch.tensor(metrics[key]).to(self.device) for key in metrics.keys()}
for key in metrics.keys():
# Create a tensor list with the same shape as the current key's tensor
tensor_list = [torch.zeros_like(torch.tensor(metrics[key])).to(self.device) for _ in range(group_size)]
# Perform the gather operation for the current key
dist.all_gather(tensor_list, metrics[key])
# Update the dictionary with the gathered tensors
tensor_dict[key] = tensor_list
mean_dict = {
key: sum(values) / len(values) if key != 'epoch' else int(values[0]) for key, values in tensor_dict.items()
}
# Convert tensors to standard Python types
mean_values_converted = {
key: value.item() if isinstance(value, torch.Tensor) else value for key, value in mean_dict.items()
}
return mean_values_converted
def train(self, output_dir, train_dataset, eval_dataset, logger, device):
self.device = device
print('self.device ', self.device)
self.logger = logger
column_names = train_dataset.column_names
model = self.pretrained_model.to(self.device)
train_dataset_samsum_pt = train_dataset.map(
self.convert_examples_to_features,
batched=True,
remove_columns=column_names,
)
eval_dataset_samsum_pt = eval_dataset.map(
self.convert_examples_to_features,
batched=True,
remove_columns=column_names,
)
# train_dataset_samsum_pt = train_dataset_samsum_pt.shard(num_shards=50, index=0) # Cut part of the train dataset to speed up testing
# eval_dataset_samsum_pt = eval_dataset_samsum_pt.shard(num_shards=50, index=0) # Cut part of the train dataset to speed up testing
seq2seq_data_collator = DataCollatorForSeq2Seq(self.tokenizer, model=model, pad_to_multiple_of=8)
train_dataloader, batch_size = self.partition_dataset(train_dataset_samsum_pt, seq2seq_data_collator)
eval_dataloader = DataLoader(eval_dataset_samsum_pt, collate_fn=seq2seq_data_collator, batch_size=batch_size)
no_decay = ["bias", "LayerNorm.weight"]
optimizer_grouped_parameters = [
{
"params": [
p for n, p in self.pretrained_model.named_parameters() if not any(nd in n for nd in no_decay)
],
"weight_decay": 0.0,
},
{
"params": [p for n, p in self.pretrained_model.named_parameters() if any(nd in n for nd in no_decay)],
"weight_decay": 0.0,
},
]
optimizer = AdamW(optimizer_grouped_parameters, lr=5e-5)
self.logger.info("***** Accelerator prepared for training *****")
num_update_steps_per_epoch = len(train_dataloader)
max_train_steps = self.num_epochs * num_update_steps_per_epoch
lr_scheduler = get_scheduler(
name='linear',
optimizer=optimizer,
num_training_steps=max_train_steps,
num_warmup_steps=1,
)
metric = evaluate.load("rouge")
# Train!
self.logger.info("***** Running training *****")
self.logger.info(f" Num examples = {len(train_dataset)}")
self.logger.info(f" Num Epochs = {self.num_epochs}")
progress_bar = tqdm(range(max_train_steps))
completed_steps = 0
for epoch in range(self.num_epochs):
model.train()
epoch_loss = 0.0
for step, batch in enumerate(train_dataloader):
outputs = model(**batch.to(self.device))
loss = outputs.loss
epoch_loss += loss.item()
loss.backward()
average_gradients(model)
optimizer.step()
lr_scheduler.step()
optimizer.zero_grad()
progress_bar.update(1)
completed_steps += 1
if completed_steps % 100 == 0:
logs = {'loss': loss.item(), 'step': completed_steps}
self.dump_valohai_metadata(logs)
if completed_steps >= max_train_steps:
break
model.eval()
gen_kwargs = {
"max_length": 128,
"num_beams": 2,
}
for step, batch in enumerate(eval_dataloader):
with torch.no_grad():
generated_tokens = self.accelerator.unwrap_model(model).generate(
batch["input_ids"].to(self.device),
attention_mask=batch["attention_mask"].to(self.device),
**gen_kwargs,
)
generated_tokens = self.accelerator.pad_across_processes(
generated_tokens,
dim=1,
pad_index=self.tokenizer.pad_token_id,
)
labels = batch["labels"]
generated_tokens = self.accelerator.gather(generated_tokens).cpu().numpy()
labels = self.accelerator.gather(labels).cpu().numpy()
labels = np.where(labels != -100, labels, self.tokenizer.pad_token_id)
if isinstance(generated_tokens, tuple):
generated_tokens = generated_tokens[0]
decoded_preds = self.tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)
decoded_labels = self.tokenizer.batch_decode(labels, skip_special_tokens=True)
# decoded_preds, decoded_labels = postprocess_text(decoded_preds, decoded_labels)
metric.add_batch(predictions=decoded_preds, references=decoded_labels)
result = metric.compute(use_stemmer=True)
# Extract a few results from ROUGE
result = {key: value * 100 for key, value in result.items()}
result = {k: round(v, 4) for k, v in result.items()}
val_logs = result.copy()
val_logs['epoch'] = epoch
avg_val_logs = self.synchronize_and_aggregate_metrics(val_logs)
print("Metrics aggregated across all machines: ")
self.dump_valohai_metadata(avg_val_logs)
self.logger.info(result)
if output_dir is not None:
self.accelerator.wait_for_everyone()
unwrapped_model = self.accelerator.unwrap_model(model)
helpers.save_valohai_metadata(unwrapped_model, output_dir)
@staticmethod
def dump_valohai_metadata(logs):
print(json.dumps(logs))
def run(my_rank, args):
print(args.output_dir)
output_dir = valohai.outputs().path(args.output_dir)
logger = logging.getLogger(__name__)
device = torch.device("cuda:{}".format(0))
data_path = os.path.dirname(valohai.inputs('dataset').path())
dataset_samsum = load_dataset(
'json',
data_files={
'train': os.path.join(data_path, 'train.json'),
'validation': os.path.join(data_path, 'val.json'),
},
)
train_dataset = dataset_samsum["train"]
eval_dataset = dataset_samsum["validation"]
print(f"Train dataset size: {len(train_dataset)}")
print(f"Test dataset size: {len(eval_dataset)}")
trainer = ModelTrainer(
model_ckpt=args.model_ckpt,
batch_size=args.batch_size,
num_epochs=args.num_epochs,
warmup_steps=args.warmup_steps,
evaluation_steps=args.evaluation_steps,
)
trainer.train(
output_dir=output_dir,
train_dataset=train_dataset,
eval_dataset=eval_dataset,
logger=logger,
device=device,
)
def init(master_url, my_rank, world_size, fn, args):
dist.init_process_group(init_method=master_url, rank=my_rank, world_size=world_size, backend='nccl')
fn(my_rank, args)
if __name__ == '__main__':
parser = argparse.ArgumentParser(description="Train a Seq2Seq model")
parser.add_argument("--model-ckpt", type=str, help="Pretrained model checkpoint")
parser.add_argument("--output-dir", type=str, help="Output directory for the trained model")
parser.add_argument("--batch-size", type=int, help="Batch size")
parser.add_argument("--num-epochs", type=int, help="Number of training epochs")
parser.add_argument("--warmup-steps", type=int, help="Warmup steps")
parser.add_argument("--evaluation-steps", type=int, help="Evaluation steps")
my_args = parser.parse_args()
master_port = 1234
master_ip = valohai.distributed.master().primary_local_ip
url = f"tcp://{master_ip}:{master_port}"
size = valohai.distributed.required_count
rank = valohai.distributed.me().rank
mp.set_start_method('spawn')
p = mp.Process(target=init, args=(url, rank, size, run, my_args))
p.start()
p.join()