输入张量大小不继承训练数据集标签计数
Posted
技术标签:
【中文标题】输入张量大小不继承训练数据集标签计数【英文标题】：Input tensor size doesnt inherit training dataset labels count 【发布时间】：2021-06-09 14:13:09 【问题描述】：最近我偶然发现了一个我自己无法解决的问题。
我一直在复制使用文本分类脚本的示例： https://github.com/huggingface/transformers/blob/master/examples/text-classification/run_glue.py https://github.com/huggingface/transformers/blob/master/examples/text-classification/run_tf_glue.py
它们的 PyTorch 和 Tensorflow 版本。
我进行了几次测试，其中一个结果我不清楚。 这些测试主要包括在两个 ML 框架脚本 (PyTorch/Tensorflow) 之间切换，以及对各自的数据集使用不同的模型。
我正在尝试使用我的自定义 CSV 数据集微调 bert-base-uncased（通过 https://huggingface.co/models 上传）等基本语言模型，以在序列分类任务中表现良好。
这些测试中的大多数都完全成功，并且我能够验证模型的推理和预测。
当我尝试在 PyTorch 框架脚本中使用不同的模型时，问题就开始了。
也许值得添加相同的模型在 Tensorflow 微调脚本上完美运行。仅当我使用 PyTorch 脚本进行微调时才会出现错误。
该模型的配置如下：
"architectures": [
"BertForPreTraining"
],
"attention_probs_dropout_prob": 0.1,
"hidden_act": "gelu",
"hidden_dropout_prob": 0.1,
"hidden_size": 768,
"initializer_range": 0.02,
"intermediate_size": 3072,
"layer_norm_eps": 1e-12,
"max_position_embeddings": 512,
"model_type": "bert",
"num_attention_heads": 12,
"num_hidden_layers": 12,
"output_past": true,
"pad_token_id": 0,
"type_vocab_size": 2,
"vocab_size": 60000
我不断收到如下错误：
RuntimeError: CUDA error: CUBLAS_STATUS_ALLOC_FAILED when calling cublasCreate(handle)
File "/home/konrad/miniconda3/envs/test_1_3/lib/python3.8/site-packages/torch/nn/functional.py", line 2387, in nll_loss
ret = torch._C._nn.nll_loss(input, target, weight, _Reduction.get_enum(reduction), ignore_index)
RuntimeError: cuda runtime error (710) : device-side assert triggered at /opt/conda/conda-bld/pytorch_1614378083779/work/aten/src/THCUNN/generic/ClassNLLCriterion.cu:115
这最终让我研究了输入和目标张量的大小。
我发现输入张量大小没有继承标签的数量。
原来是：
torch.Size([16, 2])
虽然应该是：（数据集中有 77 个唯一标签）
torch.Size([16, 77])
我很确定应该是这样，因为其他成功测试的输入张量大小是正确的。
所以我的问题出现了。 可以使用/设置什么样的方法/参数来确保正确的输入张量大小将根据我的数据集中唯一标签的数量进行调整？
BERT 模型转换器架构是否相关？ 因为当我使用像 BertModel 或 BertForMaskedLM 这样的架构模型时，没有出现类似的问题，一切都很顺利。
导致错误的实验模型具有不同的架构 - BertForPretraining。
我想问你同样的问题。 如果有任何提示和进一步的见解可以帮助我找到解决方案，我将不胜感激。
编辑：在下面添加代码。
#!/usr/bin/env python
# coding=utf-8
# Copyright 2020 The HuggingFace Inc. 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.
""" Finetuning the library models for sequence classification on GLUE."""
# You can also adapt this script on your own text classification task. Pointers for this are left as comments.
import logging
import os
import random
import sys
from dataclasses import dataclass, field
from typing import Optional
import numpy as np
from datasets import load_dataset, load_metric
import transformers
from transformers import (
AutoConfig,
BertForSequenceClassification,
AutoTokenizer,
DataCollatorWithPadding,
EvalPrediction,
HfArgumentParser,
PretrainedConfig,
Trainer,
TrainingArguments,
default_data_collator,
set_seed,
)
from transformers.trainer_utils import get_last_checkpoint, is_main_process
task_to_keys =
"cola": ("sentence", None),
"mnli": ("premise", "hypothesis"),
"mrpc": ("sentence1", "sentence2"),
"qnli": ("question", "sentence"),
"qqp": ("question1", "question2"),
"rte": ("sentence1", "sentence2"),
"sst2": ("sentence", None),
"stsb": ("sentence1", "sentence2"),
"wnli": ("sentence1", "sentence2"),
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.
"""
task_name: Optional[str] = field(
default=None,
metadata="help": "The name of the task to train on: " + ", ".join(task_to_keys.keys()),
)
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 preprocessed datasets or not."
)
pad_to_max_length: bool = field(
default=True,
metadata=
"help": "Whether to pad all samples to `max_seq_length`. "
"If False, will pad the samples dynamically when batching to the maximum length in the batch."
,
)
max_train_samples: Optional[int] = field(
default=None,
metadata=
"help": "For debugging purposes or quicker training, truncate the number of training examples to this "
"value if set."
,
)
max_val_samples: Optional[int] = field(
default=None,
metadata=
"help": "For debugging purposes or quicker training, truncate the number of validation examples to this "
"value if set."
,
)
max_test_samples: Optional[int] = field(
default=None,
metadata=
"help": "For debugging purposes or quicker training, truncate the number of test examples to this "
"value if set."
,
)
train_file: Optional[str] = field(
default=None, metadata="help": "A csv or a json file containing the training data."
)
validation_file: Optional[str] = field(
default=None, metadata="help": "A csv or a json file containing the validation data."
)
test_file: Optional[str] = field(default=None, metadata="help": "A csv or a json file containing the test data.")
def __post_init__(self):
if self.task_name is not None:
self.task_name = self.task_name.lower()
if self.task_name not in task_to_keys.keys():
raise ValueError("Unknown task, you should pick one in " + ",".join(task_to_keys.keys()))
elif self.train_file is None or self.validation_file is None:
raise ValueError("Need either a GLUE task or a training/validation file.")
else:
train_extension = self.train_file.split(".")[-1]
assert train_extension in ["csv", "json"], "`train_file` should be a csv or a json file."
validation_extension = self.validation_file.split(".")[-1]
assert (
validation_extension == train_extension
), "`validation_file` should have the same extension (csv or json) as `train_file`."
@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",
)
use_fast_tokenizer: bool = field(
default=True,
metadata="help": "Whether to use one of the fast tokenizer (backed by the tokenizers library) or not.",
)
model_revision: str = field(
default="main",
metadata="help": "The specific model version to use (can be a branch name, tag name or commit id).",
)
use_auth_token: bool = field(
default=False,
metadata=
"help": "Will use the token generated when running `transformers-cli login` (necessary to use this script "
"with private models)."
,
)
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()
# 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:
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."
)
# Setup logging
logging.basicConfig(
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
datefmt="%m/%d/%Y %H:%M:%S",
handlers=[logging.StreamHandler(sys.stdout)],
)
logger.setLevel(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(f"Training/evaluation parameters training_args")
# Set seed before initializing model.
set_seed(training_args.seed)
# Get the datasets: you can either provide your own CSV/JSON training and evaluation files (see below)
# or specify a GLUE benchmark task (the dataset will be downloaded automatically from the datasets Hub).
#
# For CSV/JSON files, this script will use as labels the column called 'label' and as pair of sentences the
# sentences in columns called 'sentence1' and 'sentence2' if such column exists or the first two columns not named
# label if at least two columns are provided.
#
# If the CSVs/JSONs contain only one non-label column, the script does single sentence classification on this
# single column. 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.task_name is not None:
# Downloading and loading a dataset from the hub.
datasets = load_dataset("glue", data_args.task_name)
else:
# Loading a dataset from your local files.
# CSV/JSON training and evaluation files are needed.
data_files = "train": data_args.train_file, "validation": data_args.validation_file
# Get the test dataset: you can provide your own CSV/JSON test file (see below)
# when you use `do_predict` without specifying a GLUE benchmark task.
if training_args.do_predict:
if data_args.test_file is not None:
train_extension = data_args.train_file.split(".")[-1]
test_extension = data_args.test_file.split(".")[-1]
assert (
test_extension == train_extension
), "`test_file` should have the same extension (csv or json) as `train_file`."
data_files["test"] = data_args.test_file
else:
raise ValueError("Need either a GLUE task or a test file for `do_predict`.")
for key in data_files.keys():
logger.info(f"load a local file for key: data_files[key]")
if data_args.train_file.endswith(".csv"):
# Loading a dataset from local csv files
datasets = load_dataset("csv", data_files=data_files)
else:
# Loading a dataset from local json files
datasets = load_dataset("json", data_files=data_files)
# See more about loading any type of standard or custom dataset at
# https://huggingface.co/docs/datasets/loading_datasets.html.
# Labels
if data_args.task_name is not None:
is_regression = data_args.task_name == "stsb"
if not is_regression:
label_list = datasets["train"].features["label"].names
num_labels = len(label_list)
else:
num_labels = 1
else:
# Trying to have good defaults here, don't hesitate to tweak to your needs.
is_regression = datasets["train"].features["label"].dtype in ["float32", "float64"]
if is_regression:
num_labels = 1
else:
# A useful fast method:
# https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.unique
label_list = datasets["train"].unique("label")
label_list.sort() # Let's sort it for determinism
num_labels = len(label_list)
print("\n\n")
print("label_list", label_list[:10])
print("\n", len(label_list))
############################################################
# Crucial addition
label2id = label: i for i, label in enumerate(label_list)
# output sorted labels to json file
import json
sorted_labels_dict = k: v for k, v in sorted(label2id.items(), key=lambda item: item[1])
with open("sorted_labels.json", "w") as outfile:
json.dump(sorted_labels_dict, outfile)
############################################################
# Load pretrained model and tokenizer
#
# In 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,
############################################################
# Crucial addition
label2id=label2id,
id2label=id: label for label, id in label2id.items(),
############################################################
finetuning_task=data_args.task_name,
cache_dir=model_args.cache_dir,
revision=model_args.model_revision,
use_auth_token=True if model_args.use_auth_token else None,
)
config.num_labels = 77
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_tokenizer,
revision=model_args.model_revision,
use_auth_token=True if model_args.use_auth_token else None,
)
model = BertForSequenceClassification.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,
revision=model_args.model_revision,
use_auth_token=True if model_args.use_auth_token else None,
)
print("\n\n\n\n", model.parameters, "\n\n\n\n")
# Preprocessing the datasets
if data_args.task_name is not None:
sentence1_key, sentence2_key = task_to_keys[data_args.task_name]
else:
# Again, we try to have some nice defaults but don't hesitate to tweak to your use case.
non_label_column_names = [name for name in datasets["train"].column_names if name != "label"]
if "sentence1" in non_label_column_names and "sentence2" in non_label_column_names:
sentence1_key, sentence2_key = "sentence1", "sentence2"
else:
if len(non_label_column_names) >= 2:
sentence1_key, sentence2_key = non_label_column_names[:2]
else:
sentence1_key, sentence2_key = non_label_column_names[0], None
# Padding strategy
if data_args.pad_to_max_length:
padding = "max_length"
else:
# We will pad later, dynamically at batch creation, to the max sequence length in each batch
padding = False
# Some models have set the order of the labels to use, so let's make sure we do use it.
label_to_id = None
if (
model.config.label2id != PretrainedConfig(num_labels=num_labels).label2id
and data_args.task_name is not None
and not is_regression
):
# Some have all caps in their config, some don't.
label_name_to_id = k.lower(): v for k, v in model.config.label2id.items()
if list(sorted(label_name_to_id.keys())) == list(sorted(label_list)):
label_to_id = i: int(label_name_to_id[label_list[i]]) for i in range(num_labels)
else:
logger.warn(
"Your model seems to have been trained with labels, but they don't match the dataset: ",
f"model labels: list(sorted(label_name_to_id.keys())), dataset labels: list(sorted(label_list))."
"\nIgnoring the model labels as a result.",
)
elif data_args.task_name is None and not is_regression:
label_to_id = v: i for i, v in enumerate(label_list)
if data_args.max_seq_length > tokenizer.model_max_length:
logger.warn(
f"The max_seq_length passed (data_args.max_seq_length) is larger than the maximum length for the"
f"model (tokenizer.model_max_length). Using max_seq_length=tokenizer.model_max_length."
)
max_seq_length = min(data_args.max_seq_length, tokenizer.model_max_length)
def preprocess_function(examples):
# Tokenize the texts
args = (
(examples[sentence1_key],) if sentence2_key is None else (examples[sentence1_key], examples[sentence2_key])
)
result = tokenizer(*args, padding=padding, max_length=max_seq_length, truncation=True)
# Map labels to IDs (not necessary for GLUE tasks)
if label_to_id is not None and "label" in examples:
result["label"] = [(label_to_id[l] if l != -1 else -1) for l in examples["label"]]
return result
if training_args.do_train:
if "train" not in datasets:
raise ValueError("--do_train requires a train dataset")
train_dataset = datasets["train"]
if data_args.max_train_samples is not None:
train_dataset = train_dataset.select(range(data_args.max_train_samples))
train_dataset = train_dataset.map(
preprocess_function,
batched=True,
load_from_cache_file=not data_args.overwrite_cache,
)
if training_args.do_eval:
if "validation" not in datasets and "validation_matched" not in datasets:
raise ValueError("--do_eval requires a validation dataset")
eval_dataset = datasets["validation_matched" if data_args.task_name == "mnli" else "validation"]
if data_args.max_val_samples is not None:
eval_dataset = eval_dataset.select(range(data_args.max_val_samples))
eval_dataset = eval_dataset.map(
preprocess_function,
batched=True,
load_from_cache_file=not data_args.overwrite_cache,
)
if training_args.do_predict or data_args.task_name is not None or data_args.test_file is not None:
if "test" not in datasets and "test_matched" not in datasets:
raise ValueError("--do_predict requires a test dataset")
test_dataset = datasets["test_matched" if data_args.task_name == "mnli" else "test"]
if data_args.max_test_samples is not None:
test_dataset = test_dataset.select(range(data_args.max_test_samples))
test_dataset = test_dataset.map(
preprocess_function,
batched=True,
load_from_cache_file=not data_args.overwrite_cache,
)
# Log a few random samples from the training set:
for index in random.sample(range(len(train_dataset)), 3):
logger.info(f"Sample index of the training set: train_dataset[index].")
# Get the metric function
if data_args.task_name is not None:
metric = load_metric("glue", data_args.task_name)
# TODO: When datasets metrics include regular accuracy, make an else here and remove special branch from
# compute_metrics
# You can define your custom compute_metrics function. It takes an `EvalPrediction` object (a namedtuple with a
# predictions and label_ids field) and has to return a dictionary string to float.
def compute_metrics(p: EvalPrediction):
preds = p.predictions[0] if isinstance(p.predictions, tuple) else p.predictions
preds = np.squeeze(preds) if is_regression else np.argmax(preds, axis=1)
if data_args.task_name is not None:
result = metric.compute(predictions=preds, references=p.label_ids)
if len(result) > 1:
result["combined_score"] = np.mean(list(result.values())).item()
return result
elif is_regression:
return "mse": ((preds - p.label_ids) ** 2).mean().item()
else:
return "accuracy": (preds == p.label_ids).astype(np.float32).mean().item()
# Data collator will default to DataCollatorWithPadding, so we change it if we already did the padding.
if data_args.pad_to_max_length:
data_collator = default_data_collator
elif training_args.fp16:
data_collator = DataCollatorWithPadding(tokenizer, pad_to_multiple_of=8)
else:
data_collator = None
# Initialize our Trainer
trainer = Trainer(
model=model,
args=training_args,
train_dataset=train_dataset,
eval_dataset=eval_dataset if training_args.do_eval else None,
compute_metrics=compute_metrics,
tokenizer=tokenizer,
data_collator=data_collator,
)
config.num_labels = 77
print("\n\n\n AutoConfig.from_pretrained(model_args.model_name_or_path).num_label\n", AutoConfig.from_pretrained(model_args.model_name_or_path).num_labels)
if training_args.do_train:
if last_checkpoint is not None:
model_path = last_checkpoint
elif os.path.isdir(model_args.model_name_or_path):
model_path = model_args.model_name_or_path
else:
model_path = None
config.num_labels = 77
train_result = trainer.train(model_path=model_path)
metrics = train_result.metrics
trainer.save_model() # Saves the tokenizer too for easy upload
output_train_file = os.path.join(training_args.output_dir, "train_results.txt")
if trainer.is_world_process_zero():
with open(output_train_file, "w") as writer:
logger.info("***** Train results *****")
for key, value in sorted(metrics.items()):
logger.info(f" key = value")
writer.write(f"key = value\n")
# Need to save the state, since Trainer.save_model saves only the tokenizer with the model
trainer.state.save_to_json(os.path.join(training_args.output_dir, "trainer_state.json"))
eval_results =
if training_args.do_eval:
logger.info("*** Evaluate ***")
# Loop to handle MNLI double evaluation (matched, mis-matched)
tasks = [data_args.task_name]
eval_datasets = [eval_dataset]
if data_args.task_name == "mnli":
tasks.append("mnli-mm")
eval_datasets.append(datasets["validation_mismatched"])
for eval_dataset, task in zip(eval_datasets, tasks):
eval_result = trainer.evaluate(eval_dataset=eval_dataset)
output_eval_file = os.path.join(training_args.output_dir, f"eval_results_task.txt")
if trainer.is_world_process_zero():
with open(output_eval_file, "w") as writer:
logger.info(f"***** Eval results task *****")
for key, value in sorted(eval_result.items()):
logger.info(f" key = value")
writer.write(f"key = value\n")
eval_results.update(eval_result)
if training_args.do_predict:
logger.info("*** Test ***")
# Loop to handle MNLI double evaluation (matched, mis-matched)
tasks = [data_args.task_name]
test_datasets = [test_dataset]
if data_args.task_name == "mnli":
tasks.append("mnli-mm")
test_datasets.append(datasets["test_mismatched"])
for test_dataset, task in zip(test_datasets, tasks):
# Removing the `label` columns because it contains -1 and Trainer won't like that.
test_dataset.remove_columns_("label")
predictions = trainer.predict(test_dataset=test_dataset).predictions
predictions = np.squeeze(predictions) if is_regression else np.argmax(predictions, axis=1)
output_test_file = os.path.join(training_args.output_dir, f"test_results_task.txt")
if trainer.is_world_process_zero():
with open(output_test_file, "w") as writer:
logger.info(f"***** Test results task *****")
writer.write("index\tprediction\n")
for index, item in enumerate(predictions):
if is_regression:
writer.write(f"index\titem:3.3f\n")
else:
item = label_list[item]
writer.write(f"index\titem\n")
def _mp_fn(index):
# For xla_spawn (TPUs)
main()
if __name__ == "__main__":
main()
【问题讨论】：
你的任务是分类任务吗？ 是的，我只关注那个。 使用 BertForSequenceCalssification 您的建议是否意味着 AutoModelForSequenceClassification 与我的模型架构（即 BertForPretraining）不匹配？ 也许值得添加相同的模型在 Tensorflow 微调脚本上完美运行。仅当我使用 PyTorch 脚本进行微调时才会出现错误。 【参考方案1】：将 transformers 更新到 4.6.1 解决了我的问题。
本期指出解决办法：https://github.com/huggingface/transformers/issues/2719。
【讨论】：
以上是关于输入张量大小不继承训练数据集标签计数的主要内容，如果未能解决你的问题，请参考以下文章