def do_train():
parser = PdArgumentParser(
(ModelArguments, DataTrainingArguments, TrainingArguments))
model_args, data_args, training_args = parser.parse_args_into_dataclasses()
paddle.set_device(training_args.device)
if paddle.distributed.get_world_size() > 1:
paddle.distributed.init_parallel_env()
# Log on each process the small summary:
logger.warning(
f"Process rank: {training_args.local_rank}, device: {training_args.device}, world_size: {training_args.world_size}, "
+
f"distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16}"
)
# 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(args)
data_args.dataset = data_args.dataset.strip()
if data_args.dataset not in ALL_DATASETS:
raise ValueError("Not found dataset {}".format(data_args.dataset))
# Use yaml config to rewrite all args.
config = ALL_DATASETS[data_args.dataset]
for args in (model_args, data_args, training_args):
for arg in vars(args):
if arg in config.keys():
setattr(args, arg, config[arg])
training_args.per_device_train_batch_size = config["batch_size"]
training_args.per_device_eval_batch_size = config["batch_size"]
dataset_config = data_args.dataset.split(" ")
raw_datasets = load_dataset(
dataset_config[0],
None if len(dataset_config) <= 1 else dataset_config[1],
)
data_args.label_list = getattr(raw_datasets['train'], "label_list", None)
num_classes = 1 if raw_datasets["train"].label_list == None else len(
raw_datasets['train'].label_list)
# Define tokenizer, model, loss function.
tokenizer = AutoTokenizer.from_pretrained(model_args.model_name_or_path)
model = AutoModelForSequenceClassification.from_pretrained(
model_args.model_name_or_path, num_classes=num_classes)
loss_fct = nn.loss.CrossEntropyLoss(
) if data_args.label_list else nn.loss.MSELoss()
# Define dataset pre-process function
if "clue" in data_args.dataset:
trans_fn = partial(clue_trans_fn, tokenizer=tokenizer, args=data_args)
else:
trans_fn = partial(seq_trans_fn, tokenizer=tokenizer, args=data_args)
# Define data collector
batchify_fn = defaut_collator(tokenizer, data_args)
# Dataset pre-process
train_dataset = raw_datasets["train"].map(trans_fn)
eval_dataset = raw_datasets["dev"].map(trans_fn)
test_dataset = raw_datasets["test"].map(trans_fn)
# Define the metrics of tasks.
def compute_metrics(p):
preds = p.predictions[0] if isinstance(p.predictions,
tuple) else p.predictions
preds = paddle.to_tensor(preds)
label = paddle.to_tensor(p.label_ids)
probs = F.softmax(preds, axis=1)
metric = Accuracy()
metric.reset()
result = metric.compute(preds, label)
metric.update(result)
accu = metric.accumulate()
metric.reset()
return {"accuracy": accu}
trainer = Trainer(
model=model,
criterion=loss_fct,
args=training_args,
data_collator=batchify_fn,
train_dataset=train_dataset,
eval_dataset=eval_dataset,
tokenizer=tokenizer,
compute_metrics=compute_metrics,
)
# Log model and data config
trainer.print_config(model_args, "Model")
trainer.print_config(data_args, "Data")
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
# Training
train_result = trainer.train(resume_from_checkpoint=checkpoint)
metrics = train_result.metrics
trainer.save_model() # Saves the tokenizer too for easy upload
trainer.log_metrics("train", metrics)
trainer.save_metrics("train", metrics)
trainer.save_state()
# Evaluate and tests model
eval_metrics = trainer.evaluate()
trainer.log_metrics("eval", eval_metrics)
test_ret = trainer.predict(test_dataset)
trainer.log_metrics("test", test_ret.metrics)
if test_ret.label_ids is None:
paddle.save(
test_ret.predictions,
os.path.join(training_args.output_dir, "test_results.pdtensor"),
)
# export inference model
input_spec = [
paddle.static.InputSpec(shape=[None, None],
dtype="int64"), # input_ids
paddle.static.InputSpec(shape=[None, None],
dtype="int64") # segment_ids
]
trainer.export_model(input_spec=input_spec,
load_best_model=True,
output_dir=model_args.export_model_dir)
def do_train():
parser = PdArgumentParser(
(ModelArguments, DataTrainingArguments, TrainingArguments))
model_args, data_args, training_args = parser.parse_args_into_dataclasses()
paddle.set_device(training_args.device)
if paddle.distributed.get_world_size() > 1:
paddle.distributed.init_parallel_env()
# Log on each process the small summary:
logger.warning(
f"Process rank: {training_args.local_rank}, device: {training_args.device}, world_size: {training_args.world_size}, "
+
f"distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16}"
)
# 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(args)
data_args.dataset = data_args.dataset.strip()
if data_args.dataset not in ALL_DATASETS:
raise ValueError("Not found dataset {}".format(data_args.dataset))
# Use yaml config to rewrite all args.
config = ALL_DATASETS[data_args.dataset]
for args in (model_args, data_args, training_args):
for arg in vars(args):
if arg in config.keys():
setattr(args, arg, config[arg])
training_args.per_device_train_batch_size = config["batch_size"]
training_args.per_device_eval_batch_size = config["batch_size"]
dataset_config = data_args.dataset.split(" ")
all_ds = load_dataset(
dataset_config[0],
None if len(dataset_config) <= 1 else dataset_config[1],
)
label_list = getattr(all_ds['train'], "label_list", None)
data_args.label_list = label_list
data_args.ignore_label = -100
data_args.no_entity_id = len(data_args.label_list) - 1
num_classes = 1 if all_ds["train"].label_list == None else len(
all_ds['train'].label_list)
# Define tokenizer, model, loss function.
tokenizer = AutoTokenizer.from_pretrained(model_args.model_name_or_path)
model = AutoModelForTokenClassification.from_pretrained(
model_args.model_name_or_path, num_classes=num_classes)
class criterion(nn.Layer):
def __init__(self):
super(criterion, self).__init__()
self.loss_fn = paddle.nn.loss.CrossEntropyLoss(
ignore_index=data_args.ignore_label)
def forward(self, *args, **kwargs):
return paddle.mean(self.loss_fn(*args, **kwargs))
loss_fct = criterion()
# Define dataset pre-process function
trans_fn = partial(ner_trans_fn, tokenizer=tokenizer, args=data_args)
# Define data collector
batchify_fn = ner_collator(tokenizer, data_args)
# Dataset pre-process
train_dataset = all_ds["train"].map(trans_fn)
eval_dataset = all_ds["dev"].map(trans_fn)
test_dataset = all_ds["test"].map(trans_fn)
# Define the metrics of tasks.
# Metrics
metric = load_metric("seqeval")
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)]
results = metric.compute(predictions=true_predictions,
references=true_labels)
return {
"precision": results["overall_precision"],
"recall": results["overall_recall"],
"f1": results["overall_f1"],
"accuracy": results["overall_accuracy"],
}
trainer = Trainer(
model=model,
criterion=loss_fct,
args=training_args,
data_collator=batchify_fn,
train_dataset=train_dataset,
eval_dataset=eval_dataset,
tokenizer=tokenizer,
compute_metrics=compute_metrics,
)
# Log model and data config
trainer.print_config(model_args, "Model")
trainer.print_config(data_args, "Data")
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
# Training
train_result = trainer.train(resume_from_checkpoint=checkpoint)
metrics = train_result.metrics
trainer.save_model() # Saves the tokenizer too for easy upload
trainer.log_metrics("train", metrics)
trainer.save_metrics("train", metrics)
trainer.save_state()
# Evaluate and tests model
eval_metrics = trainer.evaluate()
trainer.log_metrics("eval", eval_metrics)
test_ret = trainer.predict(test_dataset)
trainer.log_metrics("test", test_ret.metrics)
if test_ret.label_ids is None:
paddle.save(
test_ret.predictions,
os.path.join(training_args.output_dir, "test_results.pdtensor"),
)
# export inference model
input_spec = [
paddle.static.InputSpec(shape=[None, None],
dtype="int64"), # input_ids
paddle.static.InputSpec(shape=[None, None],
dtype="int64") # segment_ids
]
trainer.export_model(input_spec=input_spec,
load_best_model=True,
output_dir=model_args.export_model_dir)