SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

make examples consistent, revert error in num_train_steps calculation

Browse Source
This commit is contained in:
Matej Svejda
2019-01-30 11:47:25 +01:00
parent 9c6a48c8c3
commit 5169069997
5 changed files with 21 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
examples/run_classifier.py
View File

@@ -411,7 +411,7 @@ def main():
raise ValueError("Invalid gradient_accumulation_steps parameter: {}, should be >= 1".format( raise ValueError("Invalid gradient_accumulation_steps parameter: {}, should be >= 1".format(
args.gradient_accumulation_steps)) args.gradient_accumulation_steps))
args.train_batch_size = int(args.train_batch_size / args.gradient_accumulation_steps) args.train_batch_size = args.train_batch_size // args.gradient_accumulation_steps
random.seed(args.seed) random.seed(args.seed)
np.random.seed(args.seed) np.random.seed(args.seed)
@@ -441,8 +441,8 @@ def main():
num_train_steps = None num_train_steps = None
if args.do_train: if args.do_train:
train_examples = processor.get_train_examples(args.data_dir) train_examples = processor.get_train_examples(args.data_dir)
num_train_steps = int( num_train_steps =
len(train_examples) / args.train_batch_size * args.num_train_epochs) len(train_examples) // args.train_batch_size // args.gradient_accumulation_steps * args.num_train_epochs
# Prepare model # Prepare model
model = BertForSequenceClassification.from_pretrained(args.bert_model, model = BertForSequenceClassification.from_pretrained(args.bert_model,

14
examples/run_lm_finetuning.py
View File

@@ -497,7 +497,7 @@ def main():
raise ValueError("Invalid gradient_accumulation_steps parameter: {}, should be >= 1".format( raise ValueError("Invalid gradient_accumulation_steps parameter: {}, should be >= 1".format(
args.gradient_accumulation_steps)) args.gradient_accumulation_steps))
args.train_batch_size = int(args.train_batch_size / args.gradient_accumulation_steps) args.train_batch_size = args.train_batch_size // args.gradient_accumulation_steps
random.seed(args.seed) random.seed(args.seed)
np.random.seed(args.seed) np.random.seed(args.seed)
@@ -520,8 +520,8 @@ def main():
print("Loading Train Dataset", args.train_file) print("Loading Train Dataset", args.train_file)
train_dataset = BERTDataset(args.train_file, tokenizer, seq_len=args.max_seq_length, train_dataset = BERTDataset(args.train_file, tokenizer, seq_len=args.max_seq_length,
corpus_lines=None, on_memory=args.on_memory) corpus_lines=None, on_memory=args.on_memory)
num_train_steps = int( num_train_steps =
len(train_dataset) / args.train_batch_size * args.num_train_epochs) len(train_dataset) // args.train_batch_size // args.gradient_accumulation_steps * args.num_train_epochs
# Prepare model # Prepare model
model = BertForPreTraining.from_pretrained(args.bert_model) model = BertForPreTraining.from_pretrained(args.bert_model)
@@ -544,6 +544,10 @@ def main():
{'params': [p for n, p in param_optimizer if not any(nd in n for nd in no_decay)], 'weight_decay': 0.01}, {'params': [p for n, p in param_optimizer if not any(nd in n for nd in no_decay)], 'weight_decay': 0.01},
{'params': [p for n, p in param_optimizer if any(nd in n for nd in no_decay)], 'weight_decay': 0.0} {'params': [p for n, p in param_optimizer if any(nd in n for nd in no_decay)], 'weight_decay': 0.0}
] ]
t_total = num_train_steps
if args.local_rank != -1:
t_total = t_total // torch.distributed.get_world_size()
if args.fp16: if args.fp16:
try: try:
from apex.optimizers import FP16_Optimizer from apex.optimizers import FP16_Optimizer
@@ -564,7 +568,7 @@ def main():
optimizer = BertAdam(optimizer_grouped_parameters, optimizer = BertAdam(optimizer_grouped_parameters,
lr=args.learning_rate, lr=args.learning_rate,
warmup=args.warmup_proportion, warmup=args.warmup_proportion,
t_total=num_train_steps) t_total=t_total)
global_step = 0 global_step = 0
if args.do_train: if args.do_train:
@@ -604,7 +608,7 @@ def main():
if args.fp16: if args.fp16:
# modify learning rate with special warm up BERT uses # modify learning rate with special warm up BERT uses
# if args.fp16 is False, BertAdam is used that handles this automatically # if args.fp16 is False, BertAdam is used that handles this automatically
lr_this_step = args.learning_rate * warmup_linear(global_step/num_train_steps, args.warmup_proportion) lr_this_step = args.learning_rate * warmup_linear(global_step/t_total, args.warmup_proportion)
for param_group in optimizer.param_groups: for param_group in optimizer.param_groups:
param_group['lr'] = lr_this_step param_group['lr'] = lr_this_step
optimizer.step() optimizer.step()

6
examples/run_squad.py
View File

@@ -757,7 +757,7 @@ def main():
raise ValueError("Invalid gradient_accumulation_steps parameter: {}, should be >= 1".format( raise ValueError("Invalid gradient_accumulation_steps parameter: {}, should be >= 1".format(
args.gradient_accumulation_steps)) args.gradient_accumulation_steps))
args.train_batch_size = int(args.train_batch_size / args.gradient_accumulation_steps) args.train_batch_size = args.train_batch_size // args.gradient_accumulation_steps
random.seed(args.seed) random.seed(args.seed)
np.random.seed(args.seed) np.random.seed(args.seed)
@@ -788,8 +788,8 @@ def main():
if args.do_train: if args.do_train:
train_examples = read_squad_examples( train_examples = read_squad_examples(
input_file=args.train_file, is_training=True) input_file=args.train_file, is_training=True)
num_train_steps = int( num_train_steps =
len(train_examples) / args.train_batch_size * args.num_train_epochs) len(train_examples) // args.train_batch_size // args.gradient_accumulation_steps * args.num_train_epochs
# Prepare model # Prepare model
model = BertForQuestionAnswering.from_pretrained(args.bert_model, model = BertForQuestionAnswering.from_pretrained(args.bert_model,

6
examples/run_squad2.py
View File

@@ -850,7 +850,7 @@ def main():
raise ValueError("Invalid gradient_accumulation_steps parameter: {}, should be >= 1".format( raise ValueError("Invalid gradient_accumulation_steps parameter: {}, should be >= 1".format(
args.gradient_accumulation_steps)) args.gradient_accumulation_steps))
args.train_batch_size = int(args.train_batch_size / args.gradient_accumulation_steps) args.train_batch_size = args.train_batch_size // args.gradient_accumulation_steps
random.seed(args.seed) random.seed(args.seed)
np.random.seed(args.seed) np.random.seed(args.seed)
@@ -881,8 +881,8 @@ def main():
if args.do_train: if args.do_train:
train_examples = read_squad_examples( train_examples = read_squad_examples(
input_file=args.train_file, is_training=True) input_file=args.train_file, is_training=True)
num_train_steps = int( num_train_steps =
len(train_examples) / args.train_batch_size * args.num_train_epochs) len(train_examples) // args.train_batch_size // args.gradient_accumulation_steps * args.num_train_epochs
# Prepare model # Prepare model
model = BertForQuestionAnswering.from_pretrained(args.bert_model, model = BertForQuestionAnswering.from_pretrained(args.bert_model,

6
examples/run_swag.py
View File

@@ -331,7 +331,7 @@ def main():
raise ValueError("Invalid gradient_accumulation_steps parameter: {}, should be >= 1".format( raise ValueError("Invalid gradient_accumulation_steps parameter: {}, should be >= 1".format(
args.gradient_accumulation_steps)) args.gradient_accumulation_steps))
args.train_batch_size = int(args.train_batch_size / args.gradient_accumulation_steps) args.train_batch_size = args.train_batch_size // args.gradient_accumulation_steps
random.seed(args.seed) random.seed(args.seed)
np.random.seed(args.seed) np.random.seed(args.seed)
@@ -352,8 +352,8 @@ def main():
num_train_steps = None num_train_steps = None
if args.do_train: if args.do_train:
train_examples = read_swag_examples(os.path.join(args.data_dir, 'train.csv'), is_training = True) train_examples = read_swag_examples(os.path.join(args.data_dir, 'train.csv'), is_training = True)
num_train_steps = int( num_train_steps =
len(train_examples) / args.train_batch_size * args.num_train_epochs) len(train_examples) // args.train_batch_size // args.gradient_accumulation_steps * args.num_train_epochs
# Prepare model # Prepare model
model = BertForMultipleChoice.from_pretrained(args.bert_model, model = BertForMultipleChoice.from_pretrained(args.bert_model,