This commit is contained in:
2025-10-09 16:47:16 +08:00
parent c8feb4deb5
commit e27e3f16bb
5248 changed files with 1778505 additions and 0 deletions

View File

@@ -0,0 +1,243 @@
# Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team.
# Copyright (c) 2018, NVIDIA CORPORATION. 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.
"""Finetuning the library models for multiple choice (Bert, Roberta, XLNet)."""
import logging
import os
from dataclasses import dataclass, field
from typing import Optional
import numpy as np
from utils_multiple_choice import MultipleChoiceDataset, Split, processors
import transformers
from transformers import (
AutoConfig,
AutoModelForMultipleChoice,
AutoTokenizer,
DataCollatorWithPadding,
EvalPrediction,
HfArgumentParser,
Trainer,
TrainingArguments,
set_seed,
)
from transformers.trainer_utils import is_main_process
logger = logging.getLogger(__name__)
def simple_accuracy(preds, labels):
return (preds == labels).mean()
@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"}
)
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.
"""
task_name: str = field(metadata={"help": "The name of the task to train on: " + ", ".join(processors.keys())})
data_dir: str = field(metadata={"help": "Should contain the data files for the task."})
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, TrainingArguments))
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 training_args.local_rank in [-1, 0] else logging.WARN,
)
logger.warning(
"Process rank: %s, device: %s, n_gpu: %s, distributed training: %s, 16-bits training: %s",
training_args.local_rank,
training_args.device,
training_args.n_gpu,
bool(training_args.local_rank != -1),
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
set_seed(training_args.seed)
try:
processor = processors[data_args.task_name]()
label_list = processor.get_labels()
num_labels = len(label_list)
except KeyError:
raise ValueError("Task not found: %s" % (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,
)
model = AutoModelForMultipleChoice.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,
)
# Get datasets
train_dataset = (
MultipleChoiceDataset(
data_dir=data_args.data_dir,
tokenizer=tokenizer,
task=data_args.task_name,
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 = (
MultipleChoiceDataset(
data_dir=data_args.data_dir,
tokenizer=tokenizer,
task=data_args.task_name,
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 compute_metrics(p: EvalPrediction) -> dict:
preds = np.argmax(p.predictions, axis=1)
return {"acc": simple_accuracy(preds, p.label_ids)}
# Data collator
data_collator = DataCollatorWithPadding(tokenizer, pad_to_multiple_of=8) if training_args.fp16 else None
# Initialize our Trainer
trainer = Trainer(
model=model,
args=training_args,
train_dataset=train_dataset,
eval_dataset=eval_dataset,
compute_metrics=compute_metrics,
data_collator=data_collator,
)
# 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()
# For convenience, we also re-save the tokenizer to the same directory,
# so that you can share your model easily on huggingface.co/models =)
if trainer.is_world_master():
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")
if trainer.is_world_master():
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("{} = {}\n".format(key, value))
results.update(result)
return results
def _mp_fn(index):
# For xla_spawn (TPUs)
main()
if __name__ == "__main__":
main()

View File

@@ -0,0 +1,483 @@
# Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team.
# Copyright (c) 2018, NVIDIA CORPORATION. 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.
"""Multiple choice fine-tuning: utilities to work with multiple choice tasks of reading comprehension"""
import csv
import glob
import json
import logging
import os
from dataclasses import dataclass
from enum import Enum
from typing import Optional
import tqdm
from filelock import FileLock
from transformers import PreTrainedTokenizer, is_torch_available
logger = logging.getLogger(__name__)
@dataclass(frozen=True)
class InputExample:
"""
A single training/test example for multiple choice
Args:
example_id: Unique id for the example.
question: string. The untokenized text of the second sequence (question).
contexts: list of str. The untokenized text of the first sequence (context of corresponding question).
endings: list of str. multiple choice's options. Its length must be equal to contexts' length.
label: (Optional) string. The label of the example. This should be
specified for train and dev examples, but not for test examples.
"""
example_id: str
question: str
contexts: list[str]
endings: list[str]
label: Optional[str]
@dataclass(frozen=True)
class InputFeatures:
"""
A single set of features of data.
Property names are the same names as the corresponding inputs to a model.
"""
example_id: str
input_ids: list[list[int]]
attention_mask: Optional[list[list[int]]]
token_type_ids: Optional[list[list[int]]]
label: Optional[int]
class Split(Enum):
train = "train"
dev = "dev"
test = "test"
if is_torch_available():
import torch
from torch.utils.data import Dataset
class MultipleChoiceDataset(Dataset):
features: list[InputFeatures]
def __init__(
self,
data_dir: str,
tokenizer: PreTrainedTokenizer,
task: str,
max_seq_length: Optional[int] = None,
overwrite_cache=False,
mode: Split = Split.train,
):
processor = processors[task]()
cached_features_file = os.path.join(
data_dir,
"cached_{}_{}_{}_{}".format(
mode.value,
tokenizer.__class__.__name__,
str(max_seq_length),
task,
),
)
# Make sure only the first process in distributed training processes the dataset,
# and the others will use the cache.
lock_path = cached_features_file + ".lock"
with FileLock(lock_path):
if os.path.exists(cached_features_file) and not overwrite_cache:
logger.info(f"Loading features from cached file {cached_features_file}")
self.features = torch.load(cached_features_file, weights_only=True)
else:
logger.info(f"Creating features from dataset file at {data_dir}")
label_list = processor.get_labels()
if mode == Split.dev:
examples = processor.get_dev_examples(data_dir)
elif mode == Split.test:
examples = processor.get_test_examples(data_dir)
else:
examples = processor.get_train_examples(data_dir)
logger.info("Training examples: %s", len(examples))
self.features = convert_examples_to_features(
examples,
label_list,
max_seq_length,
tokenizer,
)
logger.info("Saving features into cached file %s", cached_features_file)
torch.save(self.features, cached_features_file)
def __len__(self):
return len(self.features)
def __getitem__(self, i) -> InputFeatures:
return self.features[i]
class DataProcessor:
"""Base class for data converters for multiple choice data sets."""
def get_train_examples(self, data_dir):
"""Gets a collection of `InputExample`s for the train set."""
raise NotImplementedError()
def get_dev_examples(self, data_dir):
"""Gets a collection of `InputExample`s for the dev set."""
raise NotImplementedError()
def get_test_examples(self, data_dir):
"""Gets a collection of `InputExample`s for the test set."""
raise NotImplementedError()
def get_labels(self):
"""Gets the list of labels for this data set."""
raise NotImplementedError()
class RaceProcessor(DataProcessor):
"""Processor for the RACE data set."""
def get_train_examples(self, data_dir):
"""See base class."""
logger.info(f"LOOKING AT {data_dir} train")
high = os.path.join(data_dir, "train/high")
middle = os.path.join(data_dir, "train/middle")
high = self._read_txt(high)
middle = self._read_txt(middle)
return self._create_examples(high + middle, "train")
def get_dev_examples(self, data_dir):
"""See base class."""
logger.info(f"LOOKING AT {data_dir} dev")
high = os.path.join(data_dir, "dev/high")
middle = os.path.join(data_dir, "dev/middle")
high = self._read_txt(high)
middle = self._read_txt(middle)
return self._create_examples(high + middle, "dev")
def get_test_examples(self, data_dir):
"""See base class."""
logger.info(f"LOOKING AT {data_dir} test")
high = os.path.join(data_dir, "test/high")
middle = os.path.join(data_dir, "test/middle")
high = self._read_txt(high)
middle = self._read_txt(middle)
return self._create_examples(high + middle, "test")
def get_labels(self):
"""See base class."""
return ["0", "1", "2", "3"]
def _read_txt(self, input_dir):
lines = []
files = glob.glob(input_dir + "/*txt")
for file in tqdm.tqdm(files, desc="read files"):
with open(file, encoding="utf-8") as fin:
data_raw = json.load(fin)
data_raw["race_id"] = file
lines.append(data_raw)
return lines
def _create_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for _, data_raw in enumerate(lines):
race_id = "{}-{}".format(set_type, data_raw["race_id"])
article = data_raw["article"]
for i in range(len(data_raw["answers"])):
truth = str(ord(data_raw["answers"][i]) - ord("A"))
question = data_raw["questions"][i]
options = data_raw["options"][i]
examples.append(
InputExample(
example_id=race_id,
question=question,
contexts=[article, article, article, article], # this is not efficient but convenient
endings=[options[0], options[1], options[2], options[3]],
label=truth,
)
)
return examples
class SynonymProcessor(DataProcessor):
"""Processor for the Synonym data set."""
def get_train_examples(self, data_dir):
"""See base class."""
logger.info(f"LOOKING AT {data_dir} train")
return self._create_examples(self._read_csv(os.path.join(data_dir, "mctrain.csv")), "train")
def get_dev_examples(self, data_dir):
"""See base class."""
logger.info(f"LOOKING AT {data_dir} dev")
return self._create_examples(self._read_csv(os.path.join(data_dir, "mchp.csv")), "dev")
def get_test_examples(self, data_dir):
"""See base class."""
logger.info(f"LOOKING AT {data_dir} dev")
return self._create_examples(self._read_csv(os.path.join(data_dir, "mctest.csv")), "test")
def get_labels(self):
"""See base class."""
return ["0", "1", "2", "3", "4"]
def _read_csv(self, input_file):
with open(input_file, encoding="utf-8") as f:
return list(csv.reader(f))
def _create_examples(self, lines: list[list[str]], type: str):
"""Creates examples for the training and dev sets."""
examples = [
InputExample(
example_id=line[0],
question="", # in the swag dataset, the
# common beginning of each
# choice is stored in "sent2".
contexts=[line[1], line[1], line[1], line[1], line[1]],
endings=[line[2], line[3], line[4], line[5], line[6]],
label=line[7],
)
for line in lines # we skip the line with the column names
]
return examples
class SwagProcessor(DataProcessor):
"""Processor for the SWAG data set."""
def get_train_examples(self, data_dir):
"""See base class."""
logger.info(f"LOOKING AT {data_dir} train")
return self._create_examples(self._read_csv(os.path.join(data_dir, "train.csv")), "train")
def get_dev_examples(self, data_dir):
"""See base class."""
logger.info(f"LOOKING AT {data_dir} dev")
return self._create_examples(self._read_csv(os.path.join(data_dir, "val.csv")), "dev")
def get_test_examples(self, data_dir):
"""See base class."""
logger.info(f"LOOKING AT {data_dir} dev")
raise ValueError(
"For swag testing, the input file does not contain a label column. It can not be tested in current code "
"setting!"
)
return self._create_examples(self._read_csv(os.path.join(data_dir, "test.csv")), "test")
def get_labels(self):
"""See base class."""
return ["0", "1", "2", "3"]
def _read_csv(self, input_file):
with open(input_file, encoding="utf-8") as f:
return list(csv.reader(f))
def _create_examples(self, lines: list[list[str]], type: str):
"""Creates examples for the training and dev sets."""
if type == "train" and lines[0][-1] != "label":
raise ValueError("For training, the input file must contain a label column.")
examples = [
InputExample(
example_id=line[2],
question=line[5], # in the swag dataset, the
# common beginning of each
# choice is stored in "sent2".
contexts=[line[4], line[4], line[4], line[4]],
endings=[line[7], line[8], line[9], line[10]],
label=line[11],
)
for line in lines[1:] # we skip the line with the column names
]
return examples
class ArcProcessor(DataProcessor):
"""Processor for the ARC data set (request from allennlp)."""
def get_train_examples(self, data_dir):
"""See base class."""
logger.info(f"LOOKING AT {data_dir} train")
return self._create_examples(self._read_json(os.path.join(data_dir, "train.jsonl")), "train")
def get_dev_examples(self, data_dir):
"""See base class."""
logger.info(f"LOOKING AT {data_dir} dev")
return self._create_examples(self._read_json(os.path.join(data_dir, "dev.jsonl")), "dev")
def get_test_examples(self, data_dir):
logger.info(f"LOOKING AT {data_dir} test")
return self._create_examples(self._read_json(os.path.join(data_dir, "test.jsonl")), "test")
def get_labels(self):
"""See base class."""
return ["0", "1", "2", "3"]
def _read_json(self, input_file):
with open(input_file, encoding="utf-8") as fin:
lines = fin.readlines()
return lines
def _create_examples(self, lines, type):
"""Creates examples for the training and dev sets."""
# There are two types of labels. They should be normalized
def normalize(truth):
if truth in "ABCD":
return ord(truth) - ord("A")
elif truth in "1234":
return int(truth) - 1
else:
logger.info("truth ERROR! %s", str(truth))
return None
examples = []
three_choice = 0
four_choice = 0
five_choice = 0
other_choices = 0
# we deleted example which has more than or less than four choices
for line in tqdm.tqdm(lines, desc="read arc data"):
data_raw = json.loads(line.strip("\n"))
if len(data_raw["question"]["choices"]) == 3:
three_choice += 1
continue
elif len(data_raw["question"]["choices"]) == 5:
five_choice += 1
continue
elif len(data_raw["question"]["choices"]) != 4:
other_choices += 1
continue
four_choice += 1
truth = str(normalize(data_raw["answerKey"]))
assert truth != "None"
question_choices = data_raw["question"]
question = question_choices["stem"]
id = data_raw["id"]
options = question_choices["choices"]
if len(options) == 4:
examples.append(
InputExample(
example_id=id,
question=question,
contexts=[
options[0]["para"].replace("_", ""),
options[1]["para"].replace("_", ""),
options[2]["para"].replace("_", ""),
options[3]["para"].replace("_", ""),
],
endings=[options[0]["text"], options[1]["text"], options[2]["text"], options[3]["text"]],
label=truth,
)
)
if type == "train":
assert len(examples) > 1
assert examples[0].label is not None
logger.info("len examples: %s}", str(len(examples)))
logger.info("Three choices: %s", str(three_choice))
logger.info("Five choices: %s", str(five_choice))
logger.info("Other choices: %s", str(other_choices))
logger.info("four choices: %s", str(four_choice))
return examples
def convert_examples_to_features(
examples: list[InputExample],
label_list: list[str],
max_length: int,
tokenizer: PreTrainedTokenizer,
) -> list[InputFeatures]:
"""
Loads a data file into a list of `InputFeatures`
"""
label_map = {label: i for i, label in enumerate(label_list)}
features = []
for ex_index, example in tqdm.tqdm(enumerate(examples), desc="convert examples to features"):
if ex_index % 10000 == 0:
logger.info("Writing example %d of %d" % (ex_index, len(examples)))
choices_inputs = []
for ending_idx, (context, ending) in enumerate(zip(example.contexts, example.endings)):
text_a = context
if example.question.find("_") != -1:
# this is for cloze question
text_b = example.question.replace("_", ending)
else:
text_b = example.question + " " + ending
inputs = tokenizer(
text_a,
text_b,
add_special_tokens=True,
max_length=max_length,
padding="max_length",
truncation=True,
return_overflowing_tokens=True,
)
if "num_truncated_tokens" in inputs and inputs["num_truncated_tokens"] > 0:
logger.info(
"Attention! you are cropping tokens (swag task is ok). "
"If you are training ARC and RACE and you are popping question + options, "
"you need to try to use a bigger max seq length!"
)
choices_inputs.append(inputs)
label = label_map[example.label]
input_ids = [x["input_ids"] for x in choices_inputs]
attention_mask = (
[x["attention_mask"] for x in choices_inputs] if "attention_mask" in choices_inputs[0] else None
)
token_type_ids = (
[x["token_type_ids"] for x in choices_inputs] if "token_type_ids" in choices_inputs[0] else None
)
features.append(
InputFeatures(
example_id=example.example_id,
input_ids=input_ids,
attention_mask=attention_mask,
token_type_ids=token_type_ids,
label=label,
)
)
for f in features[:2]:
logger.info("*** Example ***")
logger.info("feature: %s" % f)
return features
processors = {"race": RaceProcessor, "swag": SwagProcessor, "arc": ArcProcessor, "syn": SynonymProcessor}
MULTIPLE_CHOICE_TASKS_NUM_LABELS = {"race", 4, "swag", 4, "arc", 4, "syn", 5}