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

formating

Browse Source
This commit is contained in:
thomwolf
2020-01-06 22:55:51 +01:00
parent e25b6fe354
commit 1c9333584a
3 changed files with 353 additions and 235 deletions
Download Patch File Download Diff File Expand all files Collapse all files

111
examples/hans/hans_processors.py
View File

@@ -18,8 +18,9 @@
import logging import logging
import os import os
from utils_hans import DataProcessor, InputExample, InputFeatures
from transformers.file_utils import is_tf_available from transformers.file_utils import is_tf_available
from utils_hans import DataProcessor, InputExample, InputFeatures
if is_tf_available(): if is_tf_available():
import tensorflow as tf import tensorflow as tf
@@ -27,15 +28,18 @@ if is_tf_available():
logger = logging.getLogger(__name__) logger = logging.getLogger(__name__)
def hans_convert_examples_to_features(examples, tokenizer, def hans_convert_examples_to_features(
max_length=512, examples,
task=None, tokenizer,
label_list=None, max_length=512,
output_mode=None, task=None,
pad_on_left=False, label_list=None,
pad_token=0, output_mode=None,
pad_token_segment_id=0, pad_on_left=False,
mask_padding_with_zero=True): pad_token=0,
pad_token_segment_id=0,
mask_padding_with_zero=True,
):
""" """
Loads a data file into a list of ``InputFeatures`` Loads a data file into a list of ``InputFeatures``
@@ -82,12 +86,7 @@ def hans_convert_examples_to_features(examples, tokenizer,
example = processor.get_example_from_tensor_dict(example) example = processor.get_example_from_tensor_dict(example)
example = processor.tfds_map(example) example = processor.tfds_map(example)
inputs = tokenizer.encode_plus( inputs = tokenizer.encode_plus(example.text_a, example.text_b, add_special_tokens=True, max_length=max_length,)
example.text_a,
example.text_b,
add_special_tokens=True,
max_length=max_length,
)
input_ids, token_type_ids = inputs["input_ids"], inputs["token_type_ids"] input_ids, token_type_ids = inputs["input_ids"], inputs["token_type_ids"]
# The mask has 1 for real tokens and 0 for padding tokens. Only real # The mask has 1 for real tokens and 0 for padding tokens. Only real
@@ -106,8 +105,12 @@ def hans_convert_examples_to_features(examples, tokenizer,
token_type_ids = token_type_ids + ([pad_token_segment_id] * padding_length) token_type_ids = token_type_ids + ([pad_token_segment_id] * padding_length)
assert len(input_ids) == max_length, "Error with input length {} vs {}".format(len(input_ids), max_length) assert len(input_ids) == max_length, "Error with input length {} vs {}".format(len(input_ids), max_length)
assert len(attention_mask) == max_length, "Error with input length {} vs {}".format(len(attention_mask), max_length) assert len(attention_mask) == max_length, "Error with input length {} vs {}".format(
assert len(token_type_ids) == max_length, "Error with input length {} vs {}".format(len(token_type_ids), max_length) len(attention_mask), max_length
)
assert len(token_type_ids) == max_length, "Error with input length {} vs {}".format(
len(token_type_ids), max_length
)
if output_mode == "classification": if output_mode == "classification":
label = label_map[example.label] if example.label in label_map else 0 label = label_map[example.label] if example.label in label_map else 0
@@ -128,28 +131,40 @@ def hans_convert_examples_to_features(examples, tokenizer,
logger.info("label: %s (id = %d)" % (example.label, label)) logger.info("label: %s (id = %d)" % (example.label, label))
features.append( features.append(
InputFeatures(input_ids=input_ids, InputFeatures(
attention_mask=attention_mask, input_ids=input_ids,
token_type_ids=token_type_ids, attention_mask=attention_mask,
label=label, pairID=pairID)) token_type_ids=token_type_ids,
label=label,
pairID=pairID,
)
)
if is_tf_available() and is_tf_dataset: if is_tf_available() and is_tf_dataset:
def gen(): def gen():
for ex in features: for ex in features:
yield ({'input_ids': ex.input_ids, yield (
'attention_mask': ex.attention_mask, {
'token_type_ids': ex.token_type_ids}, "input_ids": ex.input_ids,
ex.label) "attention_mask": ex.attention_mask,
"token_type_ids": ex.token_type_ids,
},
ex.label,
)
return tf.data.Dataset.from_generator(gen, return tf.data.Dataset.from_generator(
({'input_ids': tf.int32, gen,
'attention_mask': tf.int32, ({"input_ids": tf.int32, "attention_mask": tf.int32, "token_type_ids": tf.int32}, tf.int64),
'token_type_ids': tf.int32}, (
tf.int64), {
({'input_ids': tf.TensorShape([None]), "input_ids": tf.TensorShape([None]),
'attention_mask': tf.TensorShape([None]), "attention_mask": tf.TensorShape([None]),
'token_type_ids': tf.TensorShape([None])}, "token_type_ids": tf.TensorShape([None]),
tf.TensorShape([]))) },
tf.TensorShape([]),
),
)
return features return features
@@ -159,21 +174,20 @@ class HansProcessor(DataProcessor):
def get_example_from_tensor_dict(self, tensor_dict): def get_example_from_tensor_dict(self, tensor_dict):
"""See base class.""" """See base class."""
return InputExample(tensor_dict['idx'].numpy(), return InputExample(
tensor_dict['premise'].numpy().decode('utf-8'), tensor_dict["idx"].numpy(),
tensor_dict['hypothesis'].numpy().decode('utf-8'), tensor_dict["premise"].numpy().decode("utf-8"),
str(tensor_dict['label'].numpy())) tensor_dict["hypothesis"].numpy().decode("utf-8"),
str(tensor_dict["label"].numpy()),
)
def get_train_examples(self, data_dir): def get_train_examples(self, data_dir):
"""See base class.""" """See base class."""
return self._create_examples( return self._create_examples(self._read_tsv(os.path.join(data_dir, "heuristics_train_set.txt")), "train")
self._read_tsv(os.path.join(data_dir, "heuristics_train_set.txt")), "train")
def get_dev_examples(self, data_dir): def get_dev_examples(self, data_dir):
"""See base class.""" """See base class."""
return self._create_examples( return self._create_examples(self._read_tsv(os.path.join(data_dir, "heuristics_evaluation_set.txt")), "dev")
self._read_tsv(os.path.join(data_dir, "heuristics_evaluation_set.txt")),
"dev")
def get_labels(self): def get_labels(self):
"""See base class.""" """See base class."""
@@ -188,14 +202,12 @@ class HansProcessor(DataProcessor):
guid = "%s-%s" % (set_type, line[0]) guid = "%s-%s" % (set_type, line[0])
text_a = line[5] text_a = line[5]
text_b = line[6] text_b = line[6]
pairID = line[7][2:] if line[7].startswith('ex') else line[7] pairID = line[7][2:] if line[7].startswith("ex") else line[7]
label = line[-1] label = line[-1]
examples.append( examples.append(InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label, pairID=pairID))
InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label, pairID=pairID))
return examples return examples
glue_tasks_num_labels = { glue_tasks_num_labels = {
"hans": 3, "hans": 3,
} }
@@ -207,4 +219,3 @@ glue_processors = {
glue_output_modes = { glue_output_modes = {
"hans": "classification", "hans": "classification",
} }

469
examples/hans/test_hans.py
View File

@@ -19,60 +19,72 @@ from __future__ import absolute_import, division, print_function
import argparse import argparse
import glob import glob
import json
import logging import logging
import os import os
import random import random
import json
import numpy as np import numpy as np
import torch import torch
from torch.utils.data import (DataLoader, RandomSampler, SequentialSampler, from torch.utils.data import DataLoader, RandomSampler, SequentialSampler, TensorDataset
TensorDataset)
from torch.utils.data.distributed import DistributedSampler from torch.utils.data.distributed import DistributedSampler
from tqdm import tqdm, trange
from hans_processors import glue_output_modes as output_modes
from hans_processors import glue_processors as processors
from hans_processors import hans_convert_examples_to_features as convert_examples_to_features
from transformers import (
WEIGHTS_NAME,
AdamW,
AlbertConfig,
AlbertForSequenceClassification,
AlbertTokenizer,
BertConfig,
BertForSequenceClassification,
BertTokenizer,
DistilBertConfig,
DistilBertForSequenceClassification,
DistilBertTokenizer,
RobertaConfig,
RobertaForSequenceClassification,
RobertaTokenizer,
XLMConfig,
XLMForSequenceClassification,
XLMTokenizer,
XLNetConfig,
XLNetForSequenceClassification,
XLNetTokenizer,
get_linear_schedule_with_warmup,
)
from transformers import glue_compute_metrics as compute_metrics
try: try:
from torch.utils.tensorboard import SummaryWriter from torch.utils.tensorboard import SummaryWriter
except: except:
from tensorboardX import SummaryWriter from tensorboardX import SummaryWriter
from tqdm import tqdm, trange
from transformers import (WEIGHTS_NAME, BertConfig,
BertForSequenceClassification, BertTokenizer,
RobertaConfig,
RobertaForSequenceClassification,
RobertaTokenizer,
XLMConfig, XLMForSequenceClassification,
XLMTokenizer, XLNetConfig,
XLNetForSequenceClassification,
XLNetTokenizer,
DistilBertConfig,
DistilBertForSequenceClassification,
DistilBertTokenizer,
AlbertConfig,
AlbertForSequenceClassification,
AlbertTokenizer,
)
from transformers import AdamW, get_linear_schedule_with_warmup
from transformers import glue_compute_metrics as compute_metrics
from hans_processors import glue_output_modes as output_modes
from hans_processors import glue_processors as processors
from hans_processors import hans_convert_examples_to_features as convert_examples_to_features
logger = logging.getLogger(__name__) logger = logging.getLogger(__name__)
ALL_MODELS = sum((tuple(conf.pretrained_config_archive_map.keys()) for conf in (BertConfig, XLNetConfig, XLMConfig, ALL_MODELS = sum(
RobertaConfig, DistilBertConfig)), ()) (
tuple(conf.pretrained_config_archive_map.keys())
for conf in (BertConfig, XLNetConfig, XLMConfig, RobertaConfig, DistilBertConfig)
),
(),
)
MODEL_CLASSES = { MODEL_CLASSES = {
'bert': (BertConfig, BertForSequenceClassification, BertTokenizer), "bert": (BertConfig, BertForSequenceClassification, BertTokenizer),
'xlnet': (XLNetConfig, XLNetForSequenceClassification, XLNetTokenizer), "xlnet": (XLNetConfig, XLNetForSequenceClassification, XLNetTokenizer),
'xlm': (XLMConfig, XLMForSequenceClassification, XLMTokenizer), "xlm": (XLMConfig, XLMForSequenceClassification, XLMTokenizer),
'roberta': (RobertaConfig, RobertaForSequenceClassification, RobertaTokenizer), "roberta": (RobertaConfig, RobertaForSequenceClassification, RobertaTokenizer),
'distilbert': (DistilBertConfig, DistilBertForSequenceClassification, DistilBertTokenizer), "distilbert": (DistilBertConfig, DistilBertForSequenceClassification, DistilBertTokenizer),
'albert': (AlbertConfig, AlbertForSequenceClassification, AlbertTokenizer) "albert": (AlbertConfig, AlbertForSequenceClassification, AlbertTokenizer),
} }
@@ -100,14 +112,19 @@ def train(args, train_dataset, model, tokenizer):
t_total = len(train_dataloader) // args.gradient_accumulation_steps * args.num_train_epochs t_total = len(train_dataloader) // args.gradient_accumulation_steps * args.num_train_epochs
# Prepare optimizer and schedule (linear warmup and decay) # Prepare optimizer and schedule (linear warmup and decay)
no_decay = ['bias', 'LayerNorm.weight'] no_decay = ["bias", "LayerNorm.weight"]
optimizer_grouped_parameters = [ optimizer_grouped_parameters = [
{'params': [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)], 'weight_decay': args.weight_decay}, {
{'params': [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], 'weight_decay': 0.0} "params": [p for n, p in model.named_parameters() if not any(nd in n for nd in no_decay)],
] "weight_decay": args.weight_decay,
},
{"params": [p for n, p in model.named_parameters() if any(nd in n for nd in no_decay)], "weight_decay": 0.0},
]
optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon) optimizer = AdamW(optimizer_grouped_parameters, lr=args.learning_rate, eps=args.adam_epsilon)
scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total) scheduler = get_linear_schedule_with_warmup(
optimizer, num_warmup_steps=args.warmup_steps, num_training_steps=t_total
)
if args.fp16: if args.fp16:
try: try:
from apex import amp from apex import amp
@@ -121,17 +138,21 @@ def train(args, train_dataset, model, tokenizer):
# Distributed training (should be after apex fp16 initialization) # Distributed training (should be after apex fp16 initialization)
if args.local_rank != -1: if args.local_rank != -1:
model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[args.local_rank], model = torch.nn.parallel.DistributedDataParallel(
output_device=args.local_rank, model, device_ids=[args.local_rank], output_device=args.local_rank, find_unused_parameters=True
find_unused_parameters=True) )
# Train! # Train!
logger.info("***** Running training *****") logger.info("***** Running training *****")
logger.info(" Num examples = %d", len(train_dataset)) logger.info(" Num examples = %d", len(train_dataset))
logger.info(" Num Epochs = %d", args.num_train_epochs) logger.info(" Num Epochs = %d", args.num_train_epochs)
logger.info(" Instantaneous batch size per GPU = %d", args.per_gpu_train_batch_size) logger.info(" Instantaneous batch size per GPU = %d", args.per_gpu_train_batch_size)
logger.info(" Total train batch size (w. parallel, distributed & accumulation) = %d", logger.info(
args.train_batch_size * args.gradient_accumulation_steps * (torch.distributed.get_world_size() if args.local_rank != -1 else 1)) " Total train batch size (w. parallel, distributed & accumulation) = %d",
args.train_batch_size
* args.gradient_accumulation_steps
* (torch.distributed.get_world_size() if args.local_rank != -1 else 1),
)
logger.info(" Gradient Accumulation steps = %d", args.gradient_accumulation_steps) logger.info(" Gradient Accumulation steps = %d", args.gradient_accumulation_steps)
logger.info(" Total optimization steps = %d", t_total) logger.info(" Total optimization steps = %d", t_total)
@@ -145,16 +166,16 @@ def train(args, train_dataset, model, tokenizer):
for step, batch in enumerate(epoch_iterator): for step, batch in enumerate(epoch_iterator):
model.train() model.train()
batch = tuple(t.to(args.device) for t in batch) batch = tuple(t.to(args.device) for t in batch)
inputs = {'input_ids': batch[0], inputs = {"input_ids": batch[0], "attention_mask": batch[1], "labels": batch[3]}
'attention_mask': batch[1], if args.model_type != "distilbert":
'labels': batch[3]} inputs["token_type_ids"] = (
if args.model_type != 'distilbert': batch[2] if args.model_type in ["bert", "xlnet"] else None
inputs['token_type_ids'] = batch[2] if args.model_type in ['bert', 'xlnet'] else None # XLM, DistilBERT and RoBERTa don't use segment_ids ) # XLM, DistilBERT and RoBERTa don't use segment_ids
outputs = model(**inputs) outputs = model(**inputs)
loss = outputs[0] # model outputs are always tuple in transformers (see doc) loss = outputs[0] # model outputs are always tuple in transformers (see doc)
if args.n_gpu > 1: if args.n_gpu > 1:
loss = loss.mean() # mean() to average on multi-gpu parallel training loss = loss.mean() # mean() to average on multi-gpu parallel training
if args.gradient_accumulation_steps > 1: if args.gradient_accumulation_steps > 1:
loss = loss / args.gradient_accumulation_steps loss = loss / args.gradient_accumulation_steps
@@ -178,30 +199,34 @@ def train(args, train_dataset, model, tokenizer):
if args.local_rank in [-1, 0] and args.logging_steps > 0 and global_step % args.logging_steps == 0: if args.local_rank in [-1, 0] and args.logging_steps > 0 and global_step % args.logging_steps == 0:
logs = {} logs = {}
if args.local_rank == -1 and args.evaluate_during_training: # Only evaluate when single GPU otherwise metrics may not average well if (
args.local_rank == -1 and args.evaluate_during_training
): # Only evaluate when single GPU otherwise metrics may not average well
results = evaluate(args, model, tokenizer) results = evaluate(args, model, tokenizer)
for key, value in results.items(): for key, value in results.items():
eval_key = 'eval_{}'.format(key) eval_key = "eval_{}".format(key)
logs[eval_key] = value logs[eval_key] = value
loss_scalar = (tr_loss - logging_loss) / args.logging_steps loss_scalar = (tr_loss - logging_loss) / args.logging_steps
learning_rate_scalar = scheduler.get_lr()[0] learning_rate_scalar = scheduler.get_lr()[0]
logs['learning_rate'] = learning_rate_scalar logs["learning_rate"] = learning_rate_scalar
logs['loss'] = loss_scalar logs["loss"] = loss_scalar
logging_loss = tr_loss logging_loss = tr_loss
for key, value in logs.items(): for key, value in logs.items():
tb_writer.add_scalar(key, value, global_step) tb_writer.add_scalar(key, value, global_step)
#print(json.dumps({**logs, **{'step': global_step}})) # print(json.dumps({**logs, **{'step': global_step}}))
if args.local_rank in [-1, 0] and args.save_steps > 0 and global_step % args.save_steps == 0: if args.local_rank in [-1, 0] and args.save_steps > 0 and global_step % args.save_steps == 0:
# Save model checkpoint # Save model checkpoint
output_dir = os.path.join(args.output_dir, 'checkpoint-{}'.format(global_step)) output_dir = os.path.join(args.output_dir, "checkpoint-{}".format(global_step))
if not os.path.exists(output_dir): if not os.path.exists(output_dir):
os.makedirs(output_dir) os.makedirs(output_dir)
model_to_save = model.module if hasattr(model, 'module') else model # Take care of distributed/parallel training model_to_save = (
model.module if hasattr(model, "module") else model
) # Take care of distributed/parallel training
model_to_save.save_pretrained(output_dir) model_to_save.save_pretrained(output_dir)
torch.save(args, os.path.join(output_dir, 'training_args.bin')) torch.save(args, os.path.join(output_dir, "training_args.bin"))
logger.info("Saving model checkpoint to %s", output_dir) logger.info("Saving model checkpoint to %s", output_dir)
if args.max_steps > 0 and global_step > args.max_steps: if args.max_steps > 0 and global_step > args.max_steps:
@@ -220,7 +245,7 @@ def train(args, train_dataset, model, tokenizer):
def evaluate(args, model, tokenizer, prefix=""): def evaluate(args, model, tokenizer, prefix=""):
# Loop to handle MNLI double evaluation (matched, mis-matched) # Loop to handle MNLI double evaluation (matched, mis-matched)
eval_task_names = ("mnli", "mnli-mm") if args.task_name == "mnli" else (args.task_name,) eval_task_names = ("mnli", "mnli-mm") if args.task_name == "mnli" else (args.task_name,)
eval_outputs_dirs = (args.output_dir, args.output_dir + '-MM') if args.task_name == "mnli" else (args.output_dir,) eval_outputs_dirs = (args.output_dir, args.output_dir + "-MM") if args.task_name == "mnli" else (args.output_dir,)
results = {} results = {}
for eval_task, eval_output_dir in zip(eval_task_names, eval_outputs_dirs): for eval_task, eval_output_dir in zip(eval_task_names, eval_outputs_dirs):
@@ -251,11 +276,11 @@ def evaluate(args, model, tokenizer, prefix=""):
batch = tuple(t.to(args.device) for t in batch) batch = tuple(t.to(args.device) for t in batch)
with torch.no_grad(): with torch.no_grad():
inputs = {'input_ids': batch[0], inputs = {"input_ids": batch[0], "attention_mask": batch[1], "labels": batch[3]}
'attention_mask': batch[1], if args.model_type != "distilbert":
'labels': batch[3]} inputs["token_type_ids"] = (
if args.model_type != 'distilbert': batch[2] if args.model_type in ["bert", "xlnet"] else None
inputs['token_type_ids'] = batch[2] if args.model_type in ['bert', 'xlnet'] else None # XLM, DistilBERT and RoBERTa don't use segment_ids ) # XLM, DistilBERT and RoBERTa don't use segment_ids
outputs = model(**inputs) outputs = model(**inputs)
tmp_eval_loss, logits = outputs[:2] tmp_eval_loss, logits = outputs[:2]
@@ -263,11 +288,11 @@ def evaluate(args, model, tokenizer, prefix=""):
nb_eval_steps += 1 nb_eval_steps += 1
if preds is None: if preds is None:
preds = logits.detach().cpu().numpy() preds = logits.detach().cpu().numpy()
out_label_ids = inputs['labels'].detach().cpu().numpy() out_label_ids = inputs["labels"].detach().cpu().numpy()
pair_ids = batch[4].detach().cpu().numpy() pair_ids = batch[4].detach().cpu().numpy()
else: else:
preds = np.append(preds, logits.detach().cpu().numpy(), axis=0) preds = np.append(preds, logits.detach().cpu().numpy(), axis=0)
out_label_ids = np.append(out_label_ids, inputs['labels'].detach().cpu().numpy(), axis=0) out_label_ids = np.append(out_label_ids, inputs["labels"].detach().cpu().numpy(), axis=0)
pair_ids = np.append(pair_ids, batch[4].detach().cpu().numpy(), axis=0) pair_ids = np.append(pair_ids, batch[4].detach().cpu().numpy(), axis=0)
eval_loss = eval_loss / nb_eval_steps eval_loss = eval_loss / nb_eval_steps
@@ -280,7 +305,7 @@ def evaluate(args, model, tokenizer, prefix=""):
with open(output_eval_file, "w") as writer: with open(output_eval_file, "w") as writer:
writer.write("pairID,gld_label\n") writer.write("pairID,gld_label\n")
for pid, pred in zip(pair_ids, preds): for pid, pred in zip(pair_ids, preds):
writer.write('ex' + str(pid) + ',' + label_list[int(pred)] + '\n') writer.write("ex" + str(pid) + "," + label_list[int(pred)] + "\n")
return results return results
@@ -292,11 +317,15 @@ def load_and_cache_examples(args, task, tokenizer, evaluate=False):
processor = processors[task]() processor = processors[task]()
output_mode = output_modes[task] output_mode = output_modes[task]
# Load data features from cache or dataset file # Load data features from cache or dataset file
cached_features_file = os.path.join(args.data_dir, 'cached_{}_{}_{}_{}'.format( cached_features_file = os.path.join(
'dev' if evaluate else 'train', args.data_dir,
list(filter(None, args.model_name_or_path.split('/'))).pop(), "cached_{}_{}_{}_{}".format(
str(args.max_seq_length), "dev" if evaluate else "train",
str(task))) list(filter(None, args.model_name_or_path.split("/"))).pop(),
str(args.max_seq_length),
str(task),
),
)
label_list = processor.get_labels() label_list = processor.get_labels()
@@ -305,18 +334,21 @@ def load_and_cache_examples(args, task, tokenizer, evaluate=False):
features = torch.load(cached_features_file) features = torch.load(cached_features_file)
else: else:
logger.info("Creating features from dataset file at %s", args.data_dir) logger.info("Creating features from dataset file at %s", args.data_dir)
if task in ['mnli', 'mnli-mm'] and args.model_type in ['roberta']: if task in ["mnli", "mnli-mm"] and args.model_type in ["roberta"]:
# HACK(label indices are swapped in RoBERTa pretrained model) # HACK(label indices are swapped in RoBERTa pretrained model)
label_list[1], label_list[2] = label_list[2], label_list[1] label_list[1], label_list[2] = label_list[2], label_list[1]
examples = processor.get_dev_examples(args.data_dir) if evaluate else processor.get_train_examples(args.data_dir) examples = (
features = convert_examples_to_features(examples, processor.get_dev_examples(args.data_dir) if evaluate else processor.get_train_examples(args.data_dir)
tokenizer, )
label_list=label_list, features = convert_examples_to_features(
max_length=args.max_seq_length, examples,
output_mode=output_mode, tokenizer,
pad_on_left=bool(args.model_type in ['xlnet']), # pad on the left for xlnet label_list=label_list,
pad_token=tokenizer.convert_tokens_to_ids([tokenizer.pad_token])[0], max_length=args.max_seq_length,
pad_token_segment_id=4 if args.model_type in ['xlnet'] else 0, output_mode=output_mode,
pad_on_left=bool(args.model_type in ["xlnet"]), # pad on the left for xlnet
pad_token=tokenizer.convert_tokens_to_ids([tokenizer.pad_token])[0],
pad_token_segment_id=4 if args.model_type in ["xlnet"] else 0,
) )
if args.local_rank in [-1, 0]: if args.local_rank in [-1, 0]:
logger.info("Saving features into cached file %s", cached_features_file) logger.info("Saving features into cached file %s", cached_features_file)
@@ -335,7 +367,6 @@ def load_and_cache_examples(args, task, tokenizer, evaluate=False):
all_labels = torch.tensor([f.label for f in features], dtype=torch.float) all_labels = torch.tensor([f.label for f in features], dtype=torch.float)
all_pair_ids = torch.tensor([int(f.pairID) for f in features], dtype=torch.long) all_pair_ids = torch.tensor([int(f.pairID) for f in features], dtype=torch.long)
dataset = TensorDataset(all_input_ids, all_attention_mask, all_token_type_ids, all_labels, all_pair_ids) dataset = TensorDataset(all_input_ids, all_attention_mask, all_token_type_ids, all_labels, all_pair_ids)
return dataset, label_list return dataset, label_list
@@ -344,90 +375,149 @@ def main():
parser = argparse.ArgumentParser() parser = argparse.ArgumentParser()
## Required parameters ## Required parameters
parser.add_argument("--data_dir", default=None, type=str, required=True, parser.add_argument(
help="The input data dir. Should contain the .tsv files (or other data files) for the task.") "--data_dir",
parser.add_argument("--model_type", default=None, type=str, required=True, default=None,
help="Model type selected in the list: " + ", ".join(MODEL_CLASSES.keys())) type=str,
parser.add_argument("--model_name_or_path", default=None, type=str, required=True, required=True,
help="Path to pre-trained model or shortcut name selected in the list: " + ", ".join(ALL_MODELS)) help="The input data dir. Should contain the .tsv files (or other data files) for the task.",
parser.add_argument("--task_name", default=None, type=str, required=True, )
help="The name of the task to train selected in the list: " + ", ".join(processors.keys())) parser.add_argument(
parser.add_argument("--output_dir", default=None, type=str, required=True, "--model_type",
help="The output directory where the model predictions and checkpoints will be written.") default=None,
type=str,
required=True,
help="Model type selected in the list: " + ", ".join(MODEL_CLASSES.keys()),
)
parser.add_argument(
"--model_name_or_path",
default=None,
type=str,
required=True,
help="Path to pre-trained model or shortcut name selected in the list: " + ", ".join(ALL_MODELS),
)
parser.add_argument(
"--task_name",
default=None,
type=str,
required=True,
help="The name of the task to train selected in the list: " + ", ".join(processors.keys()),
)
parser.add_argument(
"--output_dir",
default=None,
type=str,
required=True,
help="The output directory where the model predictions and checkpoints will be written.",
)
## Other parameters ## Other parameters
parser.add_argument("--config_name", default="", type=str, parser.add_argument(
help="Pretrained config name or path if not the same as model_name") "--config_name", default="", type=str, help="Pretrained config name or path if not the same as model_name"
parser.add_argument("--tokenizer_name", default="", type=str, )
help="Pretrained tokenizer name or path if not the same as model_name") parser.add_argument(
parser.add_argument("--cache_dir", default="", type=str, "--tokenizer_name",
help="Where do you want to store the pre-trained models downloaded from s3") default="",
parser.add_argument("--max_seq_length", default=128, type=int, type=str,
help="The maximum total input sequence length after tokenization. Sequences longer " help="Pretrained tokenizer name or path if not the same as model_name",
"than this will be truncated, sequences shorter will be padded.") )
parser.add_argument("--do_train", action='store_true', parser.add_argument(
help="Whether to run training.") "--cache_dir",
parser.add_argument("--do_eval", action='store_true', default="",
help="Whether to run eval on the dev set.") type=str,
parser.add_argument("--evaluate_during_training", action='store_true', help="Where do you want to store the pre-trained models downloaded from s3",
help="Rul evaluation during training at each logging step.") )
parser.add_argument("--do_lower_case", action='store_true', parser.add_argument(
help="Set this flag if you are using an uncased model.") "--max_seq_length",
default=128,
type=int,
help="The maximum total input sequence length after tokenization. Sequences longer "
"than this will be truncated, sequences shorter will be padded.",
)
parser.add_argument("--do_train", action="store_true", help="Whether to run training.")
parser.add_argument("--do_eval", action="store_true", help="Whether to run eval on the dev set.")
parser.add_argument(
"--evaluate_during_training", action="store_true", help="Rul evaluation during training at each logging step."
)
parser.add_argument(
"--do_lower_case", action="store_true", help="Set this flag if you are using an uncased model."
)
parser.add_argument("--per_gpu_train_batch_size", default=8, type=int, parser.add_argument("--per_gpu_train_batch_size", default=8, type=int, help="Batch size per GPU/CPU for training.")
help="Batch size per GPU/CPU for training.") parser.add_argument(
parser.add_argument("--per_gpu_eval_batch_size", default=8, type=int, "--per_gpu_eval_batch_size", default=8, type=int, help="Batch size per GPU/CPU for evaluation."
help="Batch size per GPU/CPU for evaluation.") )
parser.add_argument('--gradient_accumulation_steps', type=int, default=1, parser.add_argument(
help="Number of updates steps to accumulate before performing a backward/update pass.") "--gradient_accumulation_steps",
parser.add_argument("--learning_rate", default=5e-5, type=float, type=int,
help="The initial learning rate for Adam.") default=1,
parser.add_argument("--weight_decay", default=0.0, type=float, help="Number of updates steps to accumulate before performing a backward/update pass.",
help="Weight decay if we apply some.") )
parser.add_argument("--adam_epsilon", default=1e-8, type=float, parser.add_argument("--learning_rate", default=5e-5, type=float, help="The initial learning rate for Adam.")
help="Epsilon for Adam optimizer.") parser.add_argument("--weight_decay", default=0.0, type=float, help="Weight decay if we apply some.")
parser.add_argument("--max_grad_norm", default=1.0, type=float, parser.add_argument("--adam_epsilon", default=1e-8, type=float, help="Epsilon for Adam optimizer.")
help="Max gradient norm.") parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.")
parser.add_argument("--num_train_epochs", default=3.0, type=float, parser.add_argument(
help="Total number of training epochs to perform.") "--num_train_epochs", default=3.0, type=float, help="Total number of training epochs to perform."
parser.add_argument("--max_steps", default=-1, type=int, )
help="If > 0: set total number of training steps to perform. Override num_train_epochs.") parser.add_argument(
parser.add_argument("--warmup_steps", default=0, type=int, "--max_steps",
help="Linear warmup over warmup_steps.") default=-1,
type=int,
help="If > 0: set total number of training steps to perform. Override num_train_epochs.",
)
parser.add_argument("--warmup_steps", default=0, type=int, help="Linear warmup over warmup_steps.")
parser.add_argument('--logging_steps', type=int, default=50, parser.add_argument("--logging_steps", type=int, default=50, help="Log every X updates steps.")
help="Log every X updates steps.") parser.add_argument("--save_steps", type=int, default=50, help="Save checkpoint every X updates steps.")
parser.add_argument('--save_steps', type=int, default=50, parser.add_argument(
help="Save checkpoint every X updates steps.") "--eval_all_checkpoints",
parser.add_argument("--eval_all_checkpoints", action='store_true', action="store_true",
help="Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number") help="Evaluate all checkpoints starting with the same prefix as model_name ending and ending with step number",
parser.add_argument("--no_cuda", action='store_true', )
help="Avoid using CUDA when available") parser.add_argument("--no_cuda", action="store_true", help="Avoid using CUDA when available")
parser.add_argument('--overwrite_output_dir', action='store_true', parser.add_argument(
help="Overwrite the content of the output directory") "--overwrite_output_dir", action="store_true", help="Overwrite the content of the output directory"
parser.add_argument('--overwrite_cache', action='store_true', )
help="Overwrite the cached training and evaluation sets") parser.add_argument(
parser.add_argument('--seed', type=int, default=42, "--overwrite_cache", action="store_true", help="Overwrite the cached training and evaluation sets"
help="random seed for initialization") )
parser.add_argument("--seed", type=int, default=42, help="random seed for initialization")
parser.add_argument('--fp16', action='store_true', parser.add_argument(
help="Whether to use 16-bit (mixed) precision (through NVIDIA apex) instead of 32-bit") "--fp16",
parser.add_argument('--fp16_opt_level', type=str, default='O1', action="store_true",
help="For fp16: Apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3']." help="Whether to use 16-bit (mixed) precision (through NVIDIA apex) instead of 32-bit",
"See details at https://nvidia.github.io/apex/amp.html") )
parser.add_argument("--local_rank", type=int, default=-1, parser.add_argument(
help="For distributed training: local_rank") "--fp16_opt_level",
parser.add_argument('--server_ip', type=str, default='', help="For distant debugging.") type=str,
parser.add_argument('--server_port', type=str, default='', help="For distant debugging.") default="O1",
help="For fp16: Apex AMP optimization level selected in ['O0', 'O1', 'O2', and 'O3']."
"See details at https://nvidia.github.io/apex/amp.html",
)
parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")
parser.add_argument("--server_ip", type=str, default="", help="For distant debugging.")
parser.add_argument("--server_port", type=str, default="", help="For distant debugging.")
args = parser.parse_args() args = parser.parse_args()
if os.path.exists(args.output_dir) and os.listdir(args.output_dir) and args.do_train and not args.overwrite_output_dir: if (
raise ValueError("Output directory ({}) already exists and is not empty. Use --overwrite_output_dir to overcome.".format(args.output_dir)) os.path.exists(args.output_dir)
and os.listdir(args.output_dir)
and args.do_train
and not args.overwrite_output_dir
):
raise ValueError(
"Output directory ({}) already exists and is not empty. Use --overwrite_output_dir to overcome.".format(
args.output_dir
)
)
# Setup distant debugging if needed # Setup distant debugging if needed
if args.server_ip and args.server_port: if args.server_ip and args.server_port:
# Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script # Distant debugging - see https://code.visualstudio.com/docs/python/debugging#_attach-to-a-local-script
import ptvsd import ptvsd
print("Waiting for debugger attach") print("Waiting for debugger attach")
ptvsd.enable_attach(address=(args.server_ip, args.server_port), redirect_output=True) ptvsd.enable_attach(address=(args.server_ip, args.server_port), redirect_output=True)
ptvsd.wait_for_attach() ptvsd.wait_for_attach()
@@ -439,16 +529,24 @@ def main():
else: # Initializes the distributed backend which will take care of sychronizing nodes/GPUs else: # Initializes the distributed backend which will take care of sychronizing nodes/GPUs
torch.cuda.set_device(args.local_rank) torch.cuda.set_device(args.local_rank)
device = torch.device("cuda", args.local_rank) device = torch.device("cuda", args.local_rank)
torch.distributed.init_process_group(backend='nccl') torch.distributed.init_process_group(backend="nccl")
args.n_gpu = 1 args.n_gpu = 1
args.device = device args.device = device
# Setup logging # Setup logging
logging.basicConfig(format = '%(asctime)s - %(levelname)s - %(name)s - %(message)s', logging.basicConfig(
datefmt = '%m/%d/%Y %H:%M:%S', format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
level = logging.INFO if args.local_rank in [-1, 0] else logging.WARN) datefmt="%m/%d/%Y %H:%M:%S",
logger.warning("Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s", level=logging.INFO if args.local_rank in [-1, 0] else logging.WARN,
args.local_rank, device, args.n_gpu, bool(args.local_rank != -1), args.fp16) )
logger.warning(
"Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s",
args.local_rank,
device,
args.n_gpu,
bool(args.local_rank != -1),
args.fp16,
)
# Set seed # Set seed
set_seed(args) set_seed(args)
@@ -468,17 +566,23 @@ def main():
args.model_type = args.model_type.lower() args.model_type = args.model_type.lower()
config_class, model_class, tokenizer_class = MODEL_CLASSES[args.model_type] config_class, model_class, tokenizer_class = MODEL_CLASSES[args.model_type]
config = config_class.from_pretrained(args.config_name if args.config_name else args.model_name_or_path, config = config_class.from_pretrained(
num_labels=num_labels, args.config_name if args.config_name else args.model_name_or_path,
finetuning_task=args.task_name, num_labels=num_labels,
cache_dir=args.cache_dir if args.cache_dir else None) finetuning_task=args.task_name,
tokenizer = tokenizer_class.from_pretrained(args.tokenizer_name if args.tokenizer_name else args.model_name_or_path, cache_dir=args.cache_dir if args.cache_dir else None,
do_lower_case=args.do_lower_case, )
cache_dir=args.cache_dir if args.cache_dir else None) tokenizer = tokenizer_class.from_pretrained(
model = model_class.from_pretrained(args.model_name_or_path, args.tokenizer_name if args.tokenizer_name else args.model_name_or_path,
from_tf=bool('.ckpt' in args.model_name_or_path), do_lower_case=args.do_lower_case,
config=config, cache_dir=args.cache_dir if args.cache_dir else None,
cache_dir=args.cache_dir if args.cache_dir else None) )
model = model_class.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,
)
if args.local_rank == 0: if args.local_rank == 0:
torch.distributed.barrier() # Make sure only the first process in distributed training will download model & vocab torch.distributed.barrier() # Make sure only the first process in distributed training will download model & vocab
@@ -487,14 +591,12 @@ def main():
logger.info("Training/evaluation parameters %s", args) logger.info("Training/evaluation parameters %s", args)
# Training # Training
if args.do_train: if args.do_train:
train_dataset, _ = load_and_cache_examples(args, args.task_name, tokenizer, evaluate=False) train_dataset, _ = load_and_cache_examples(args, args.task_name, tokenizer, evaluate=False)
global_step, tr_loss = train(args, train_dataset, model, tokenizer) global_step, tr_loss = train(args, train_dataset, model, tokenizer)
logger.info(" global_step = %s, average loss = %s", global_step, tr_loss) logger.info(" global_step = %s, average loss = %s", global_step, tr_loss)
# Saving best-practices: if you use defaults names for the model, you can reload it using from_pretrained() # Saving best-practices: if you use defaults names for the model, you can reload it using from_pretrained()
if args.do_train and (args.local_rank == -1 or torch.distributed.get_rank() == 0): if args.do_train and (args.local_rank == -1 or torch.distributed.get_rank() == 0):
# Create output directory if needed # Create output directory if needed
@@ -504,36 +606,39 @@ def main():
logger.info("Saving model checkpoint to %s", args.output_dir) logger.info("Saving model checkpoint to %s", args.output_dir)
# Save a trained model, configuration and tokenizer using `save_pretrained()`. # Save a trained model, configuration and tokenizer using `save_pretrained()`.
# They can then be reloaded using `from_pretrained()` # They can then be reloaded using `from_pretrained()`
model_to_save = model.module if hasattr(model, 'module') else model # Take care of distributed/parallel training model_to_save = (
model.module if hasattr(model, "module") else model
) # Take care of distributed/parallel training
model_to_save.save_pretrained(args.output_dir) model_to_save.save_pretrained(args.output_dir)
tokenizer.save_pretrained(args.output_dir) tokenizer.save_pretrained(args.output_dir)
# Good practice: save your training arguments together with the trained model # Good practice: save your training arguments together with the trained model
torch.save(args, os.path.join(args.output_dir, 'training_args.bin')) torch.save(args, os.path.join(args.output_dir, "training_args.bin"))
# Load a trained model and vocabulary that you have fine-tuned # Load a trained model and vocabulary that you have fine-tuned
model = model_class.from_pretrained(args.output_dir) model = model_class.from_pretrained(args.output_dir)
tokenizer = tokenizer_class.from_pretrained(args.output_dir) tokenizer = tokenizer_class.from_pretrained(args.output_dir)
model.to(args.device) model.to(args.device)
# Evaluation # Evaluation
results = {} results = {}
if args.do_eval and args.local_rank in [-1, 0]: if args.do_eval and args.local_rank in [-1, 0]:
tokenizer = tokenizer_class.from_pretrained(args.output_dir, do_lower_case=args.do_lower_case) tokenizer = tokenizer_class.from_pretrained(args.output_dir, do_lower_case=args.do_lower_case)
checkpoints = [args.output_dir] checkpoints = [args.output_dir]
if args.eval_all_checkpoints: if args.eval_all_checkpoints:
checkpoints = list(os.path.dirname(c) for c in sorted(glob.glob(args.output_dir + '/**/' + WEIGHTS_NAME, recursive=True))) checkpoints = list(
os.path.dirname(c) for c in sorted(glob.glob(args.output_dir + "/**/" + WEIGHTS_NAME, recursive=True))
)
logging.getLogger("transformers.modeling_utils").setLevel(logging.WARN) # Reduce logging logging.getLogger("transformers.modeling_utils").setLevel(logging.WARN) # Reduce logging
logger.info("Evaluate the following checkpoints: %s", checkpoints) logger.info("Evaluate the following checkpoints: %s", checkpoints)
for checkpoint in checkpoints: for checkpoint in checkpoints:
global_step = checkpoint.split('-')[-1] if len(checkpoints) > 1 else "" global_step = checkpoint.split("-")[-1] if len(checkpoints) > 1 else ""
prefix = checkpoint.split('/')[-1] if checkpoint.find('checkpoint') != -1 else "" prefix = checkpoint.split("/")[-1] if checkpoint.find("checkpoint") != -1 else ""
model = model_class.from_pretrained(checkpoint) model = model_class.from_pretrained(checkpoint)
model.to(args.device) model.to(args.device)
result = evaluate(args, model, tokenizer, prefix=prefix) result = evaluate(args, model, tokenizer, prefix=prefix)
result = dict((k + '_{}'.format(global_step), v) for k, v in result.items()) result = dict((k + "_{}".format(global_step), v) for k, v in result.items())
results.update(result) results.update(result)
return results return results

8
examples/hans/utils_hans.py
View File

@@ -14,10 +14,11 @@
# See the License for the specific language governing permissions and # See the License for the specific language governing permissions and
# limitations under the License. # limitations under the License.
import csv
import sys
import copy import copy
import csv
import json import json
import sys
class InputExample(object): class InputExample(object):
""" """
@@ -32,6 +33,7 @@ class InputExample(object):
label: (Optional) string. The label of the example. This should be label: (Optional) string. The label of the example. This should be
specified for train and dev examples, but not for test examples. specified for train and dev examples, but not for test examples.
""" """
def __init__(self, guid, text_a, text_b=None, label=None, pairID=None): def __init__(self, guid, text_a, text_b=None, label=None, pairID=None):
self.guid = guid self.guid = guid
self.text_a = text_a self.text_a = text_a
@@ -117,6 +119,6 @@ class DataProcessor(object):
lines = [] lines = []
for line in reader: for line in reader:
if sys.version_info[0] == 2: if sys.version_info[0] == 2:
line = list(unicode(cell, 'utf-8') for cell in line) line = list(unicode(cell, "utf-8") for cell in line)
lines.append(line) lines.append(line)
return lines return lines