Skip to content

GitLab

Unverified Commit 73a53265 authored by Matt's avatar Matt Committed by GitHub
Browse files

New TF GLUE example (#12028) 



* Pushing partially-complete new GLUE example

* First draft of the new TF GLUE example! Needs a little more testing to be sure but it's almost ready.

* Fix to the fit() call

* Bugfixes, making sure TPU and multi-GPU support is ready

* Remove logger line that depends on Pytorch

* Style pass

* Deleting old TF GLUE example

* Include label2id and id2label in the saved model config

* Don't clobber the existing model.config.label2id

* Style fixes

* Update examples/tensorflow/text-classification/run_glue.py
Co-authored-by: default avatarSylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: default avatarSylvain Gugger <35901082+sgugger@users.noreply.github.com>
parent 9d2cee8b
Expand all Show whitespace changes
Inline Side-by-side
Showing with 558 additions and 269 deletions
examples/tensorflow/text-classification/run_glue.py 0 → 100644
View file @ 73a53265
This diff is collapsed.
examples/tensorflow/text-classification/run_text_classification.py
View file @ 73a53265
......@@ -205,7 +205,6 @@ class ModelArguments:
"with private models)."
},
)
tpu: Optional[str] = field(default=None, metadata={"help": "Name of the TPU resource to use, if available"})
# endregion
......@@ -439,10 +438,8 @@ def main():
model.compile(optimizer=optimizer, loss=loss_fn, metrics=metrics)
# endregion
# region Convert data to TF format
# region Convert data to a tf.data.Dataset
# Convert data to a tf.keras.utils.Sequence object for training if we're not using a TPU
# For TPU, convert to a tf.data.Dataset
tf_data = dict()
max_samples = {
"train": data_args.max_train_samples,
......
examples/tensorflow/text-classification/run_tf_glue.py deleted 100755 → 0
View file @ 9d2cee8b
#!/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 enum import Enum
from typing import Dict, Optional
import numpy as np
import tensorflow as tf
import tensorflow_datasets as tfds
from transformers import (
AutoConfig,
AutoTokenizer,
EvalPrediction,
HfArgumentParser,
PreTrainedTokenizer,
TFAutoModelForSequenceClassification,
TFTrainer,
TFTrainingArguments,
glue_compute_metrics,
glue_convert_examples_to_features,
glue_output_modes,
glue_processors,
glue_tasks_num_labels,
)
from transformers.utils import logging as hf_logging
hf_logging.set_verbosity_info()
hf_logging.enable_default_handler()
hf_logging.enable_explicit_format()
class Split(Enum):
train = "train"
dev = "validation"
test = "test"
def get_tfds(
task_name: str,
tokenizer: PreTrainedTokenizer,
max_seq_length: Optional[int] = None,
mode: Split = Split.train,
data_dir: str = None,
):
if task_name == "mnli-mm" and mode == Split.dev:
tfds_name = "mnli_mismatched"
elif task_name == "mnli-mm" and mode == Split.train:
tfds_name = "mnli"
elif task_name == "mnli" and mode == Split.dev:
tfds_name = "mnli_matched"
elif task_name == "sst-2":
tfds_name = "sst2"
elif task_name == "sts-b":
tfds_name = "stsb"
else:
tfds_name = task_name
ds, info = tfds.load("glue/" + tfds_name, split=mode.value, with_info=True, data_dir=data_dir)
ds = glue_convert_examples_to_features(ds, tokenizer, max_seq_length, task_name)
ds = ds.apply(tf.data.experimental.assert_cardinality(info.splits[mode.value].num_examples))
return ds
logger = logging.getLogger(__name__)
@dataclass
class GlueDataTrainingArguments:
"""
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.
"""
task_name: str = field(metadata={"help": "The name of the task to train on: " + ", ".join(glue_processors.keys())})
data_dir: Optional[str] = field(default=None, metadata={"help": "The input/output data dir for TFDS."})
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 __post_init__(self):
self.task_name = self.task_name.lower()
@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, GlueDataTrainingArguments, 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}")
try:
num_labels = glue_tasks_num_labels["mnli" if data_args.task_name == "mnli-mm" else data_args.task_name]
output_mode = glue_output_modes[data_args.task_name]
except KeyError:
raise ValueError(f"Task not found: {data_args.task_name}")
# 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,
)
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,
)
# Get datasets
train_dataset = (
get_tfds(
task_name=data_args.task_name,
tokenizer=tokenizer,
max_seq_length=data_args.max_seq_length,
data_dir=data_args.data_dir,
)
if training_args.do_train
else None
)
eval_dataset = (
get_tfds(
task_name=data_args.task_name,
tokenizer=tokenizer,
max_seq_length=data_args.max_seq_length,
mode=Split.dev,
data_dir=data_args.data_dir,
)
if training_args.do_eval
else None
)
def compute_metrics(p: EvalPrediction) -> Dict:
if output_mode == "classification":
preds = np.argmax(p.predictions, axis=1)
elif output_mode == "regression":
preds = np.squeeze(p.predictions)
return glue_compute_metrics(data_args.task_name, preds, p.label_ids)
# 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()
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment