def test_trainer_iterable_dataset(self):
MODEL_ID = "sshleifer/tiny-distilbert-base-cased"
model = AutoModelForSequenceClassification.from_pretrained(MODEL_ID)
train_dataset = SampleIterableDataset(PATH_SAMPLE_TEXT)
training_args = TrainingArguments(output_dir="./examples",
no_cuda=True)
trainer = Trainer(model=model,
args=training_args,
train_dataset=train_dataset)
loader = trainer.get_train_dataloader()
self.assertIsInstance(loader, torch.utils.data.DataLoader)
def test_data_is_not_parallelized_when_model_is_parallel(self):
model = RegressionModel()
# Make the Trainer believe it's a parallelized model
model.is_parallelizable = True
model.model_parallel = True
args = TrainingArguments("./regression",
per_device_train_batch_size=16,
per_device_eval_batch_size=16)
trainer = Trainer(model,
args,
train_dataset=RegressionDataset(),
eval_dataset=RegressionDataset())
# Check the Trainer was fooled
self.assertTrue(trainer.is_model_parallel)
self.assertEqual(trainer.args.n_gpu, 1)
# The batch size of the training and evaluation dataloaders should be 16, not 16 * n_gpu
self.assertEqual(trainer.get_train_dataloader().batch_size, 16)
self.assertEqual(len(trainer.get_train_dataloader()), 64 // 16)
self.assertEqual(trainer.get_eval_dataloader().batch_size, 16)
self.assertEqual(len(trainer.get_eval_dataloader()), 64 // 16)
def test_training_iterable_dataset(self):
config = RegressionModelConfig()
model = RegressionPreTrainedModel(config)
train_dataset = SampleIterableDataset()
args = RegressionTrainingArguments(output_dir="./examples", max_steps=4)
trainer = Trainer(model=model, args=args, train_dataset=train_dataset)
trainer.train()
self.assertEqual(trainer.state.global_step, 4)
loader = trainer.get_train_dataloader()
self.assertIsInstance(loader, torch.utils.data.DataLoader)
self.assertIsInstance(loader.sampler, torch.utils.data.dataloader._InfiniteConstantSampler)
def test_trainer_iterable_dataset(self):
# Simulate Language Modeling with an IterableDataset, with no __len__ method
# Pick-up a tiny model, so it works on CPU
# See Issue #5990: https://github.com/huggingface/transformers/issues/5990
MODEL_ID = "sshleifer/tiny-distilbert-base-cased"
model = AutoModelForMaskedLM.from_pretrained(MODEL_ID)
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
train_dataset = SampleIterableDataset(file_path=PATH_SAMPLE_TEXT, tokenizer=tokenizer)
training_args = TrainingArguments(output_dir="./examples", no_cuda=True, max_steps=2)
data_collator = DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=True, mlm_probability=0.15)
training_args = TrainingArguments(output_dir="./examples", no_cuda=True, max_steps=2)
trainer = Trainer(model=model, args=training_args, train_dataset=train_dataset, data_collator=data_collator)
trainer.train()
loader = trainer.get_train_dataloader()
self.assertIsInstance(loader, torch.utils.data.DataLoader)
self.assertIsInstance(loader.sampler, torch.utils.data.dataloader._InfiniteConstantSampler)
# Exception if giving iterable dataset and no max_steps
with self.assertRaises(ValueError):
training_args = TrainingArguments(output_dir="./examples", no_cuda=True)
_ = Trainer(model=model, args=training_args, train_dataset=train_dataset, data_collator=data_collator)
# Exception if eval_dataset is iterable in __init__
with self.assertRaises(ValueError):
training_args = TrainingArguments(output_dir="./examples", no_cuda=True, max_steps=2)
_ = Trainer(
model=model,
args=training_args,
train_dataset=train_dataset,
eval_dataset=train_dataset,
data_collator=data_collator,
)
# Exception if predicting with iterable dataset
with self.assertRaises(ValueError):
training_args = TrainingArguments(output_dir="./examples", no_cuda=True)
trainer = Trainer(model=model, args=training_args, data_collator=data_collator)
trainer.predict(train_dataset)
# Exception if evaluating with iterable dataset
with self.assertRaises(ValueError):
training_args = TrainingArguments(output_dir="./examples", no_cuda=True)
trainer = Trainer(model=model, args=training_args, data_collator=data_collator)
trainer.evaluate(train_dataset)
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, 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(
)
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.")
# Setup logging
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()
transformers.utils.logging.enable_default_handler()
transformers.utils.logging.enable_explicit_format()
logger.info("Training/evaluation parameters %s", training_args)
# Set seed before initializing model.
set_seed(training_args.seed)
# 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.
# **YD** force the load_dataset performs locally
raise ValueError('MUST use local dataset!')
# datasets = load_dataset(data_args.dataset_name, data_args.dataset_config_name)
else:
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["validation"] = data_args.validation_file
if data_args.test_file is not None:
data_files["test"] = data_args.test_file
extension = data_args.extension_file
print('extension', extension)
print('data_files', data_files)
datasets = 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.
if training_args.do_train:
column_names = datasets["train"].column_names
features = datasets["train"].features
elif training_args.do_eval:
column_names = datasets["validation"].column_names
features = datasets["validation"].features
elif training_args.do_predict:
column_names = datasets["test"].column_names
features = datasets["test"].features
else:
raise ValueError('must do_train/do_eval/do_predict')
text_column_name = "tokens" if "tokens" in column_names else column_names[0]
label_column_name = (f"{data_args.task_name}_tags"
if f"{data_args.task_name}_tags" in column_names else
column_names[1])
# In the event the labels are not a `Sequence[ClassLabel]`, we will need to go through the dataset to get the
# unique labels.
def get_label_list(labels):
unique_labels = set()
for label in labels:
unique_labels = unique_labels | set(label)
label_list = list(unique_labels)
label_list.sort()
return label_list
if isinstance(features[label_column_name].feature, ClassLabel):
label_list = features[label_column_name].feature.names
# No need to convert the labels since they are already ints.
label_to_id = {i: i for i in range(len(label_list))}
else:
label_list = get_label_list(datasets["train"][label_column_name])
label_to_id = {l: i for i, l in enumerate(label_list)}
num_labels = len(label_list)
# Load pretrained model and tokenizer
#
# Distributed training:
# The .from_pretrained methods guarantee that only one local process can concurrently
# download model & vocab.
config = AutoConfig.from_pretrained(
model_args.config_name
if model_args.config_name else model_args.model_name_or_path,
num_labels=num_labels,
finetuning_task=data_args.task_name,
cache_dir=model_args.cache_dir,
)
tokenizer = AutoTokenizer.from_pretrained(
model_args.tokenizer_name
if model_args.tokenizer_name else model_args.model_name_or_path,
cache_dir=model_args.cache_dir,
use_fast=True,
)
model = AutoModelForTokenClassification.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,
)
# Preprocessing the dataset
# Padding strategy
padding = "max_length" if data_args.pad_to_max_length else False
# Tokenize all texts and align the labels with them.
def tokenize_and_align_labels(examples):
tokenized_inputs = tokenizer(
examples[text_column_name],
padding=padding,
truncation=True,
# We use this argument because the texts in our dataset are lists of words (with a label for each word).
is_split_into_words=True,
return_offsets_mapping=True,
)
offset_mappings = tokenized_inputs.pop("offset_mapping")
labels = []
for label, offset_mapping in zip(examples[label_column_name],
offset_mappings):
label_index = 0
current_label = -100
label_ids = []
for offset in offset_mapping:
# We set the label for the first token of each word. Special characters will have an offset of (0, 0)
# so the test ignores them.
if offset[0] == 0 and offset[1] != 0:
current_label = label_to_id[label[label_index]]
label_index += 1
label_ids.append(current_label)
# For special tokens, we set the label to -100 so it's automatically ignored in the loss function.
elif offset[0] == 0 and offset[1] == 0:
label_ids.append(-100)
# For the other tokens in a word, we set the label to either the current label or -100, depending on
# the label_all_tokens flag.
else:
label_ids.append(
current_label if data_args.label_all_tokens else -100)
labels.append(label_ids)
tokenized_inputs["labels"] = labels
return tokenized_inputs
tokenized_datasets = datasets.map(
tokenize_and_align_labels,
batched=True,
num_proc=data_args.preprocessing_num_workers,
load_from_cache_file=not data_args.overwrite_cache,
)
# Data collator
data_collator = DataCollatorForTokenClassification(tokenizer)
print(tokenized_datasets['train'][0])
train_dataset = tokenized_datasets['train']
tmp_columns = ['attention_mask', 'input_ids', 'labels', 'token_type_ids']
train_dataset.set_format(type=train_dataset.format["type"],
columns=tmp_columns)
data_loader = DataLoader(
tokenized_datasets['train'],
batch_size=8,
# sampler=RandomSampler(tokenized_datasets['train']),
sampler=None,
collate_fn=data_collator,
drop_last=False,
num_workers=0,
)
for i, input in enumerate(data_loader):
print('i', i)
print(input)
sys.exit()
# Metrics
def compute_metrics(p):
predictions, labels = p
predictions = np.argmax(predictions, axis=2)
# Remove ignored index (special tokens)
true_predictions = [[
label_list[p] for (p, l) in zip(prediction, label) if l != -100
] for prediction, label in zip(predictions, labels)]
true_labels = [[
label_list[l] for (p, l) in zip(prediction, label) if l != -100
] for prediction, label in zip(predictions, labels)]
return {
"accuracy_score": accuracy_score(true_labels, true_predictions),
"precision": precision_score(true_labels, true_predictions),
"recall": recall_score(true_labels, true_predictions),
"f1": f1_score(true_labels, true_predictions),
}
# Initialize our Trainer
trainer = Trainer(
model=model,
args=training_args,
train_dataset=tokenized_datasets["train"]
if training_args.do_train else None,
eval_dataset=tokenized_datasets["validation"]
if training_args.do_eval else None,
tokenizer=tokenizer,
data_collator=data_collator,
compute_metrics=compute_metrics,
)
train_dataloader = trainer.get_train_dataloader()
print('train_dataloader', list(train_dataloader)[0])
print('train_dataloader: dataset', train_dataloader.dataset.column_names)
sys.exit()
# Training
if training_args.do_train:
trainer.train(model_path=model_args.model_name_or_path if os.path.
isdir(model_args.model_name_or_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_ner.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")
# Predict
if training_args.do_predict:
logger.info("*** Predict ***")
test_dataset = tokenized_datasets["test"]
predictions, labels, metrics = trainer.predict(test_dataset)
predictions = np.argmax(predictions, axis=2)
# Remove ignored index (special tokens)
true_predictions = [[
label_list[p] for (p, l) in zip(prediction, label) if l != -100
] for prediction, label in zip(predictions, labels)]
output_test_results_file = os.path.join(training_args.output_dir,
"test_results.txt")
if trainer.is_world_process_zero():
with open(output_test_results_file, "w") as writer:
for key, value in metrics.items():
logger.info(f" {key} = {value}")
writer.write(f"{key} = {value}\n")
# Save predictions
output_test_predictions_file = os.path.join(training_args.output_dir,
"test_predictions.txt")
if trainer.is_world_process_zero():
with open(output_test_predictions_file, "w") as writer:
for prediction in true_predictions:
writer.write(" ".join(prediction) + "\n")
return results
def main(config):
# get relevant tweets
my_tweets = dl_tweets(config.handle)
curated_tweets = [fix_text(tweet) for tweet in my_tweets]
clean_tweets = [clean_tweet(tweet) for tweet in curated_tweets]
cool_tweets = [tweet for tweet in clean_tweets if not boring_tweet(tweet)]
# split train/validation sets
random.shuffle(cool_tweets)
split_train_valid = 0.9
train_size = int(split_train_valid * len(cool_tweets))
valid_size = len(cool_tweets) - train_size
train_dataset, valid_dataset = torch.utils.data.random_split(
cool_tweets, [train_size, valid_size])
# make data files
with open('data_{}_train.txt'.format(config.handle), 'w') as f:
data = make_dataset(train_dataset, config.epochs)
f.write(data)
with open('data_{}_valid.txt'.format(config.handle), 'w') as f:
data = make_dataset(valid_dataset, 1)
f.write(data)
# Set up training parameters
tokenizer = AutoTokenizer.from_pretrained('gpt2')
model = AutoModelForCausalLM.from_pretrained('gpt2')
block_size = tokenizer.max_len
train_dataset = TextDataset(tokenizer=tokenizer,
file_path=f'data_{config.handle}_train.txt',
block_size=block_size,
overwrite_cache=True)
valid_dataset = TextDataset(tokenizer=tokenizer,
file_path=f'data_{config.handle}_valid.txt',
block_size=block_size,
overwrite_cache=True)
data_collator = DataCollatorForLanguageModeling(tokenizer=tokenizer,
mlm=False)
seed = random.randint(0, 2**32 - 1)
training_args = TrainingArguments(
output_dir=f'output/{config.handle}',
overwrite_output_dir=True,
do_train=True,
do_eval=True,
evaluate_during_training=True,
num_train_epochs=1,
per_device_train_batch_size=1,
logging_steps=5,
eval_steps=5,
save_steps=0,
learning_rate=config.learning_rate,
gradient_accumulation_steps=config.gradient_accumulation_steps,
seed=seed)
os.environ['WANDB_WATCH'] = 'false' # used in Trainer
trainer = Trainer(model=model,
args=training_args,
data_collator=data_collator,
train_dataset=train_dataset,
eval_dataset=valid_dataset,
prediction_loss_only=True)
# Update lr scheduler
train_dataloader = trainer.get_train_dataloader()
num_train_steps = int(
len(train_dataloader) // config.gradient_accumulation_steps)
optimizer, _ = trainer.get_optimizers(num_train_steps)
if config.lr_scheduler == 'constant':
scheduler = get_constant_schedule_with_warmup(
optimizer,
num_warmup_steps=int(config.percent_warmup_steps *
num_train_steps))
elif config.lr_scheduler == 'linear':
scheduler = get_linear_schedule_with_warmup(
optimizer,
num_warmup_steps=int(config.percent_warmup_steps *
num_train_steps),
num_training_steps=num_train_steps)
elif config.lr_scheduler == 'cosine':
scheduler = get_cosine_schedule_with_warmup(
optimizer,
num_warmup_steps=int(config.percent_warmup_steps *
num_train_steps),
num_training_steps=num_train_steps)
trainer.optimizers = (optimizer, scheduler)
# Train & evaluate
trainer.train()
trainer.evaluate()
