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

TF: purge TFTrainer (#28483)

Browse Source
This commit is contained in:
Joao Gante
2024-01-12 16:56:34 +00:00
committed by GitHub
parent afc45b13ca
commit 4fb3d3a0f6
15 changed files with 233 additions and 1682 deletions
Download Patch File Download Diff File Expand all files Collapse all files

313
examples/legacy/text-classification/run_tf_text_classification.py
View File

@@ -1,313 +0,0 @@
#!/usr/bin/env python
# coding=utf-8
# Copyright 2020 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
""" Fine-tuning the library models for sequence classification."""
import logging
import os
from dataclasses import dataclass, field
from typing import Dict, Optional
import datasets
import numpy as np
import tensorflow as tf
from transformers import (
AutoConfig,
AutoTokenizer,
EvalPrediction,
HfArgumentParser,
PreTrainedTokenizer,
TFAutoModelForSequenceClassification,
TFTrainer,
TFTrainingArguments,
)
from transformers.utils import logging as hf_logging
hf_logging.set_verbosity_info()
hf_logging.enable_default_handler()
hf_logging.enable_explicit_format()
def get_tfds(
train_file: str,
eval_file: str,
test_file: str,
tokenizer: PreTrainedTokenizer,
label_column_id: int,
max_seq_length: Optional[int] = None,
):
files = {}
if train_file is not None:
files[datasets.Split.TRAIN] = [train_file]
if eval_file is not None:
files[datasets.Split.VALIDATION] = [eval_file]
if test_file is not None:
files[datasets.Split.TEST] = [test_file]
ds = datasets.load_dataset("csv", data_files=files)
features_name = list(ds[list(files.keys())[0]].features.keys())
label_name = features_name.pop(label_column_id)
label_list = list(set(ds[list(files.keys())[0]][label_name]))
label2id = {label: i for i, label in enumerate(label_list)}
input_names = tokenizer.model_input_names
transformed_ds = {}
if len(features_name) == 1:
for k in files.keys():
transformed_ds[k] = ds[k].map(
lambda example: tokenizer.batch_encode_plus(
example[features_name[0]], truncation=True, max_length=max_seq_length, padding="max_length"
),
batched=True,
)
elif len(features_name) == 2:
for k in files.keys():
transformed_ds[k] = ds[k].map(
lambda example: tokenizer.batch_encode_plus(
(example[features_name[0]], example[features_name[1]]),
truncation=True,
max_length=max_seq_length,
padding="max_length",
),
batched=True,
)
def gen_train():
for ex in transformed_ds[datasets.Split.TRAIN]:
d = {k: v for k, v in ex.items() if k in input_names}
label = label2id[ex[label_name]]
yield (d, label)
def gen_val():
for ex in transformed_ds[datasets.Split.VALIDATION]:
d = {k: v for k, v in ex.items() if k in input_names}
label = label2id[ex[label_name]]
yield (d, label)
def gen_test():
for ex in transformed_ds[datasets.Split.TEST]:
d = {k: v for k, v in ex.items() if k in input_names}
label = label2id[ex[label_name]]
yield (d, label)
train_ds = (
tf.data.Dataset.from_generator(
gen_train,
({k: tf.int32 for k in input_names}, tf.int64),
({k: tf.TensorShape([None]) for k in input_names}, tf.TensorShape([])),
)
if datasets.Split.TRAIN in transformed_ds
else None
)
if train_ds is not None:
train_ds = train_ds.apply(tf.data.experimental.assert_cardinality(len(ds[datasets.Split.TRAIN])))
val_ds = (
tf.data.Dataset.from_generator(
gen_val,
({k: tf.int32 for k in input_names}, tf.int64),
({k: tf.TensorShape([None]) for k in input_names}, tf.TensorShape([])),
)
if datasets.Split.VALIDATION in transformed_ds
else None
)
if val_ds is not None:
val_ds = val_ds.apply(tf.data.experimental.assert_cardinality(len(ds[datasets.Split.VALIDATION])))
test_ds = (
tf.data.Dataset.from_generator(
gen_test,
({k: tf.int32 for k in input_names}, tf.int64),
({k: tf.TensorShape([None]) for k in input_names}, tf.TensorShape([])),
)
if datasets.Split.TEST in transformed_ds
else None
)
if test_ds is not None:
test_ds = test_ds.apply(tf.data.experimental.assert_cardinality(len(ds[datasets.Split.TEST])))
return train_ds, val_ds, test_ds, label2id
logger = logging.getLogger(__name__)
@dataclass
class DataTrainingArguments:
"""
Arguments pertaining to what data we are going to input our model for training and eval.
Using `HfArgumentParser` we can turn this class
into argparse arguments to be able to specify them on
the command line.
"""
label_column_id: int = field(metadata={"help": "Which column contains the label"})
train_file: str = field(default=None, metadata={"help": "The path of the training file"})
dev_file: Optional[str] = field(default=None, metadata={"help": "The path of the development file"})
test_file: Optional[str] = field(default=None, metadata={"help": "The path of the test file"})
max_seq_length: int = field(
default=128,
metadata={
"help": (
"The maximum total input sequence length after tokenization. Sequences longer "
"than this will be truncated, sequences shorter will be padded."
)
},
)
overwrite_cache: bool = field(
default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
)
@dataclass
class ModelArguments:
"""
Arguments pertaining to which model/config/tokenizer we are going to fine-tune from.
"""
model_name_or_path: str = field(
metadata={"help": "Path to pretrained model or model identifier from huggingface.co/models"}
)
config_name: Optional[str] = field(
default=None, metadata={"help": "Pretrained config name or path if not the same as model_name"}
)
tokenizer_name: Optional[str] = field(
default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
)
use_fast: bool = field(default=False, metadata={"help": "Set this flag to use fast tokenization."})
# If you want to tweak more attributes on your tokenizer, you should do it in a distinct script,
# or just modify its tokenizer_config.json.
cache_dir: Optional[str] = field(
default=None,
metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
)
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, TFTrainingArguments))
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,
)
logger.info(
f"n_replicas: {training_args.n_replicas}, distributed training: {bool(training_args.n_replicas > 1)}, "
f"16-bits training: {training_args.fp16}"
)
logger.info(f"Training/evaluation parameters {training_args}")
# Load pretrained model and tokenizer
#
# Distributed training:
# The .from_pretrained methods guarantee that only one local process can concurrently
# download model & vocab.
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,
)
train_dataset, eval_dataset, test_ds, label2id = get_tfds(
train_file=data_args.train_file,
eval_file=data_args.dev_file,
test_file=data_args.test_file,
tokenizer=tokenizer,
label_column_id=data_args.label_column_id,
max_seq_length=data_args.max_seq_length,
)
config = AutoConfig.from_pretrained(
model_args.config_name if model_args.config_name else model_args.model_name_or_path,
num_labels=len(label2id),
label2id=label2id,
id2label={id: label for label, id in label2id.items()},
finetuning_task="text-classification",
cache_dir=model_args.cache_dir,
)
with training_args.strategy.scope():
model = TFAutoModelForSequenceClassification.from_pretrained(
model_args.model_name_or_path,
from_pt=bool(".bin" in model_args.model_name_or_path),
config=config,
cache_dir=model_args.cache_dir,
)
def compute_metrics(p: EvalPrediction) -> Dict:
preds = np.argmax(p.predictions, axis=1)
return {"acc": (preds == p.label_ids).mean()}
# Initialize our Trainer
trainer = TFTrainer(
model=model,
args=training_args,
train_dataset=train_dataset,
eval_dataset=eval_dataset,
compute_metrics=compute_metrics,
)
# Training
if training_args.do_train:
trainer.train()
trainer.save_model()
tokenizer.save_pretrained(training_args.output_dir)
# Evaluation
results = {}
if training_args.do_eval:
logger.info("*** Evaluate ***")
result = trainer.evaluate()
output_eval_file = os.path.join(training_args.output_dir, "eval_results.txt")
with open(output_eval_file, "w") as writer:
logger.info("***** Eval results *****")
for key, value in result.items():
logger.info(f" {key} = {value}")
writer.write(f"{key} = {value}\n")
results.update(result)
return results
if __name__ == "__main__":
main()

310
examples/legacy/token-classification/run_tf_ner.py
View File

@@ -1,310 +0,0 @@
#!/usr/bin/env python
# coding=utf-8
# Copyright 2018 The HuggingFace Inc. team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
""" Fine-tuning the library models for named entity recognition."""
import logging
import os
from dataclasses import dataclass, field
from importlib import import_module
from typing import Dict, List, Optional, Tuple
import numpy as np
from seqeval.metrics import classification_report, f1_score, precision_score, recall_score
from utils_ner import Split, TFTokenClassificationDataset, TokenClassificationTask
from transformers import (
AutoConfig,
AutoTokenizer,
EvalPrediction,
HfArgumentParser,
TFAutoModelForTokenClassification,
TFTrainer,
TFTrainingArguments,
)
from transformers.utils import logging as hf_logging
hf_logging.set_verbosity_info()
hf_logging.enable_default_handler()
hf_logging.enable_explicit_format()
logger = logging.getLogger(__name__)
@dataclass
class ModelArguments:
"""
Arguments pertaining to which model/config/tokenizer we are going to fine-tune from.
"""
model_name_or_path: str = field(
metadata={"help": "Path to pretrained model or model identifier from huggingface.co/models"}
)
config_name: Optional[str] = field(
default=None, metadata={"help": "Pretrained config name or path if not the same as model_name"}
)
task_type: Optional[str] = field(
default="NER", metadata={"help": "Task type to fine tune in training (e.g. NER, POS, etc)"}
)
tokenizer_name: Optional[str] = field(
default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
)
use_fast: bool = field(default=False, metadata={"help": "Set this flag to use fast tokenization."})
# If you want to tweak more attributes on your tokenizer, you should do it in a distinct script,
# or just modify its tokenizer_config.json.
cache_dir: Optional[str] = field(
default=None,
metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
)
@dataclass
class DataTrainingArguments:
"""
Arguments pertaining to what data we are going to input our model for training and eval.
"""
data_dir: str = field(
metadata={"help": "The input data dir. Should contain the .txt files for a CoNLL-2003-formatted task."}
)
labels: Optional[str] = field(
metadata={"help": "Path to a file containing all labels. If not specified, CoNLL-2003 labels are used."}
)
max_seq_length: int = field(
default=128,
metadata={
"help": (
"The maximum total input sequence length after tokenization. Sequences longer "
"than this will be truncated, sequences shorter will be padded."
)
},
)
overwrite_cache: bool = field(
default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
)
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, TFTrainingArguments))
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."
)
module = import_module("tasks")
try:
token_classification_task_clazz = getattr(module, model_args.task_type)
token_classification_task: TokenClassificationTask = token_classification_task_clazz()
except AttributeError:
raise ValueError(
f"Task {model_args.task_type} needs to be defined as a TokenClassificationTask subclass in {module}. "
f"Available tasks classes are: {TokenClassificationTask.__subclasses__()}"
)
# Setup logging
logging.basicConfig(
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
datefmt="%m/%d/%Y %H:%M:%S",
level=logging.INFO,
)
logger.info(
"n_replicas: %s, distributed training: %s, 16-bits training: %s",
training_args.n_replicas,
bool(training_args.n_replicas > 1),
training_args.fp16,
)
logger.info("Training/evaluation parameters %s", training_args)
# Prepare Token Classification task
labels = token_classification_task.get_labels(data_args.labels)
label_map: Dict[int, str] = dict(enumerate(labels))
num_labels = len(labels)
# 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,
id2label=label_map,
label2id={label: i for i, label in enumerate(labels)},
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=model_args.use_fast,
)
with training_args.strategy.scope():
model = TFAutoModelForTokenClassification.from_pretrained(
model_args.model_name_or_path,
from_pt=bool(".bin" in model_args.model_name_or_path),
config=config,
cache_dir=model_args.cache_dir,
)
# Get datasets
train_dataset = (
TFTokenClassificationDataset(
token_classification_task=token_classification_task,
data_dir=data_args.data_dir,
tokenizer=tokenizer,
labels=labels,
model_type=config.model_type,
max_seq_length=data_args.max_seq_length,
overwrite_cache=data_args.overwrite_cache,
mode=Split.train,
)
if training_args.do_train
else None
)
eval_dataset = (
TFTokenClassificationDataset(
token_classification_task=token_classification_task,
data_dir=data_args.data_dir,
tokenizer=tokenizer,
labels=labels,
model_type=config.model_type,
max_seq_length=data_args.max_seq_length,
overwrite_cache=data_args.overwrite_cache,
mode=Split.dev,
)
if training_args.do_eval
else None
)
def align_predictions(predictions: np.ndarray, label_ids: np.ndarray) -> Tuple[List[int], List[int]]:
preds = np.argmax(predictions, axis=2)
batch_size, seq_len = preds.shape
out_label_list = [[] for _ in range(batch_size)]
preds_list = [[] for _ in range(batch_size)]
for i in range(batch_size):
for j in range(seq_len):
if label_ids[i, j] != -100:
out_label_list[i].append(label_map[label_ids[i][j]])
preds_list[i].append(label_map[preds[i][j]])
return preds_list, out_label_list
def compute_metrics(p: EvalPrediction) -> Dict:
preds_list, out_label_list = align_predictions(p.predictions, p.label_ids)
return {
"precision": precision_score(out_label_list, preds_list),
"recall": recall_score(out_label_list, preds_list),
"f1": f1_score(out_label_list, preds_list),
}
# Initialize our Trainer
trainer = TFTrainer(
model=model,
args=training_args,
train_dataset=train_dataset.get_dataset() if train_dataset else None,
eval_dataset=eval_dataset.get_dataset() if eval_dataset else None,
compute_metrics=compute_metrics,
)
# Training
if training_args.do_train:
trainer.train()
trainer.save_model()
tokenizer.save_pretrained(training_args.output_dir)
# Evaluation
results = {}
if training_args.do_eval:
logger.info("*** Evaluate ***")
result = trainer.evaluate()
output_eval_file = os.path.join(training_args.output_dir, "eval_results.txt")
with open(output_eval_file, "w") as writer:
logger.info("***** Eval results *****")
for key, value in result.items():
logger.info(" %s = %s", key, value)
writer.write("%s = %s\n" % (key, value))
results.update(result)
# Predict
if training_args.do_predict:
test_dataset = TFTokenClassificationDataset(
token_classification_task=token_classification_task,
data_dir=data_args.data_dir,
tokenizer=tokenizer,
labels=labels,
model_type=config.model_type,
max_seq_length=data_args.max_seq_length,
overwrite_cache=data_args.overwrite_cache,
mode=Split.test,
)
predictions, label_ids, metrics = trainer.predict(test_dataset.get_dataset())
preds_list, labels_list = align_predictions(predictions, label_ids)
report = classification_report(labels_list, preds_list)
logger.info("\n%s", report)
output_test_results_file = os.path.join(training_args.output_dir, "test_results.txt")
with open(output_test_results_file, "w") as writer:
writer.write("%s\n" % report)
# Save predictions
output_test_predictions_file = os.path.join(training_args.output_dir, "test_predictions.txt")
with open(output_test_predictions_file, "w") as writer:
with open(os.path.join(data_args.data_dir, "test.txt"), "r") as f:
example_id = 0
for line in f:
if line.startswith("-DOCSTART-") or line == "" or line == "\n":
writer.write(line)
if not preds_list[example_id]:
example_id += 1
elif preds_list[example_id]:
output_line = line.split()[0] + " " + preds_list[example_id].pop(0) + "\n"
writer.write(output_line)
else:
logger.warning("Maximum sequence length exceeded: No prediction for '%s'.", line.split()[0])
return results
if __name__ == "__main__":
main()

4
examples/pytorch/README.md
View File

@@ -226,7 +226,7 @@ wandb.login()
To enable logging to W&B, include `"wandb"` in the `report_to` of your `TrainingArguments` or script. Or just pass along `--report_to_all` if you have `wandb` installed.
Whenever you use `Trainer` or `TFTrainer` classes, your losses, evaluation metrics, model topology and gradients (for `Trainer` only) will automatically be logged.
Whenever you use the `Trainer` class, your losses, evaluation metrics, model topology and gradients will automatically be logged.
Advanced configuration is possible by setting environment variables:
@@ -282,7 +282,7 @@ To enable Neptune logging, in your `TrainingArguments`, set the `report_to` argu
```python
training_args = TrainingArguments(
"quick-training-distilbert-mrpc",
"quick-training-distilbert-mrpc",
evaluation_strategy="steps",
eval_steps=20,
report_to="neptune",

8
examples/tensorflow/README.md
View File

@@ -15,7 +15,7 @@ limitations under the License.
# Examples
This folder contains actively maintained examples of the use of 🤗 Transformers organized into different ML tasks. All examples in this folder are **TensorFlow** examples and are written using native Keras rather than classes like `TFTrainer`, which we now consider deprecated. If you've previously only used 🤗 Transformers via `TFTrainer`, we highly recommend taking a look at the new style - we think it's a big improvement!
This folder contains actively maintained examples of the use of 🤗 Transformers organized into different ML tasks. All examples in this folder are **TensorFlow** examples and are written using native Keras. If you've previously only used 🤗 Transformers via `TFTrainer`, we highly recommend taking a look at the new style - we think it's a big improvement!
In addition, all scripts here now support the [🤗 Datasets](https://github.com/huggingface/datasets) library - you can grab entire datasets just by changing one command-line argument!
@@ -32,13 +32,13 @@ Here is the list of all our examples:
| Task | Example datasets |
|---|---|
| [**`language-modeling`**](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/language-modeling) | WikiText-2
| [**`multiple-choice`**](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/multiple-choice) | SWAG
| [**`multiple-choice`**](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/multiple-choice) | SWAG
| [**`question-answering`**](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/question-answering) | SQuAD
| [**`summarization`**](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/summarization) | XSum
| [**`summarization`**](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/summarization) | XSum
| [**`text-classification`**](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/text-classification) | GLUE
| [**`token-classification`**](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/token-classification) | CoNLL NER
| [**`translation`**](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/translation) | WMT
## Coming soon
- **Colab notebooks** to easily run through these scripts!
- **Colab notebooks** to easily run through these scripts!