南京建站公司网站,什么是网络营销中最古老的一种,网站备案一次吗,互联网商城建设我们知道bert-base的大小大约在400M左右#xff0c;有时候我们的任务比较简单#xff0c;并不需要如此重量级的bert#xff0c;这时候#xff0c;我们可以使用轻量级的tiny-bert#xff08;100M以内#xff09;#xff0c;在保证性能的同时#xff0c;降低对硬件的门槛…我们知道bert-base的大小大约在400M左右有时候我们的任务比较简单并不需要如此重量级的bert这时候我们可以使用轻量级的tiny-bert100M以内在保证性能的同时降低对硬件的门槛。
本博客主要介绍
1. 预训练数据集
2. 预训练代码 一. 数据集
魔搭社区
数据集我是用的上面的这个链接数据量很大每个文件都有1G大家可以随便挑选其中的部分进行训练本博客我只使用了1个约1.3G的数据进行训练 二. 预训练代码 #!/usr/bin/env python
# codingutf-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 masked language modeling (BERT, ALBERT, RoBERTa...) on a text file or a dataset.Here is the full list of checkpoints on the hub that can be fine-tuned by this script:
https://huggingface.co/models?filterfill-mask# You can also adapt this script on your own masked language modeling task. Pointers for this are left as comments.import logging
import math
import os
import sys
import warnings
from dataclasses import dataclass, field
from itertools import chain
from typing import Optionalimport datasets
import evaluate
import torch
from datasets import load_datasetimport transformers
from transformers import (CONFIG_MAPPING,MODEL_FOR_MASKED_LM_MAPPING,AutoConfig,AutoModelForMaskedLM,AutoTokenizer,DataCollatorForLanguageModeling,HfArgumentParser,Trainer,TrainingArguments,is_torch_xla_available,set_seed,
)
from transformers.trainer_utils import get_last_checkpoint
from transformers.utils import check_min_version, send_example_telemetry
from transformers.utils.versions import require_version# Will error if the minimal version of Transformers is not installed. Remove at your own risks.
check_min_version(4.40.0.dev0)require_version(datasets1.8.0, To fix: pip install -r examples/pytorch/language-modeling/requirements.txt)logger logging.getLogger(__name__)
MODEL_CONFIG_CLASSES list(MODEL_FOR_MASKED_LM_MAPPING.keys())
MODEL_TYPES tuple(conf.model_type for conf in MODEL_CONFIG_CLASSES)dataclass
class ModelArguments:Arguments pertaining to which model/config/tokenizer we are going to fine-tune, or train from scratch.model_name_or_path: Optional[str] field(defaultNone,metadata{help: (The model checkpoint for weights initialization. Dont set if you want to train a model from scratch.)},)model_type: Optional[str] field(defaultNone,metadata{help: If training from scratch, pass a model type from the list: , .join(MODEL_TYPES)},)config_overrides: Optional[str] field(defaultNone,metadata{help: (Override some existing default config settings when a model is trained from scratch. Example: n_embd10,resid_pdrop0.2,scale_attn_weightsfalse,summary_typecls_index)},)config_name: Optional[str] field(defaultNone, metadata{help: Pretrained config name or path if not the same as model_name})tokenizer_name: Optional[str] field(defaultNone, metadata{help: Pretrained tokenizer name or path if not the same as model_name})cache_dir: Optional[str] field(defaultNone,metadata{help: Where do you want to store the pretrained models downloaded from huggingface.co},)use_fast_tokenizer: bool field(defaultTrue,metadata{help: Whether to use one of the fast tokenizer (backed by the tokenizers library) or not.},)model_revision: str field(defaultmain,metadata{help: The specific model version to use (can be a branch name, tag name or commit id).},)token: str field(defaultNone,metadata{help: (The token to use as HTTP bearer authorization for remote files. If not specified, will use the token generated when running huggingface-cli login (stored in ~/.huggingface).)},)use_auth_token: bool field(defaultNone,metadata{help: The use_auth_token argument is deprecated and will be removed in v4.34. Please use token instead.},)trust_remote_code: bool field(defaultFalse,metadata{help: (Whether or not to allow for custom models defined on the Hub in their own modeling files. This option should only be set to True for repositories you trust and in which you have read the code, as it will execute code present on the Hub on your local machine.)},)torch_dtype: Optional[str] field(defaultNone,metadata{help: (Override the default torch.dtype and load the model under this dtype. If auto is passed, the dtype will be automatically derived from the models weights.),choices: [auto, bfloat16, float16, float32],},)low_cpu_mem_usage: bool field(defaultFalse,metadata{help: (It is an option to create the model as an empty shell, then only materialize its parameters when the pretrained weights are loaded. set True will benefit LLM loading time and RAM consumption.)},)def __post_init__(self):if self.config_overrides is not None and (self.config_name is not None or self.model_name_or_path is not None):raise ValueError(--config_overrides cant be used in combination with --config_name or --model_name_or_path)dataclass
class DataTrainingArguments:Arguments pertaining to what data we are going to input our model for training and eval.dataset_name: Optional[str] field(defaultNone, metadata{help: The name of the dataset to use (via the datasets library).})dataset_config_name: Optional[str] field(defaultNone, metadata{help: The configuration name of the dataset to use (via the datasets library).})train_file: Optional[str] field(defaultNone, metadata{help: The input training data file (a text file).})validation_file: Optional[str] field(defaultNone,metadata{help: An optional input evaluation data file to evaluate the perplexity on (a text file).},)overwrite_cache: bool field(defaultFalse, metadata{help: Overwrite the cached training and evaluation sets})validation_split_percentage: Optional[int] field(default5,metadata{help: The percentage of the train set used as validation set in case theres no validation split},)max_seq_length: Optional[int] field(defaultNone,metadata{help: (The maximum total input sequence length after tokenization. Sequences longer than this will be truncated.)},)preprocessing_num_workers: Optional[int] field(defaultNone,metadata{help: The number of processes to use for the preprocessing.},)mlm_probability: float field(default0.15, metadata{help: Ratio of tokens to mask for masked language modeling loss})line_by_line: bool field(defaultFalse,metadata{help: Whether distinct lines of text in the dataset are to be handled as distinct sequences.},)pad_to_max_length: bool field(defaultFalse,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(defaultNone,metadata{help: (For debugging purposes or quicker training, truncate the number of training examples to this value if set.)},)max_eval_samples: Optional[int] field(defaultNone,metadata{help: (For debugging purposes or quicker training, truncate the number of evaluation examples to this value if set.)},)streaming: bool field(defaultFalse, metadata{help: Enable streaming mode})def __post_init__(self):if self.streaming:require_version(datasets2.0.0, The streaming feature requires datasets2.0.0)if self.dataset_name is None and self.train_file is None and self.validation_file is None:raise ValueError(Need either a dataset name or a training/validation file.)else:if self.train_file is not None:extension self.train_file.split(.)[-1]if extension not in [csv, json, txt]:raise ValueError(train_file should be a csv, a json or a txt file.)if self.validation_file is not None:extension self.validation_file.split(.)[-1]if extension not in [csv, json, txt]:raise ValueError(validation_file should be a csv, a json or a txt file.)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 its the path to a json file,# lets parse it to get our arguments.model_args, data_args, training_args parser.parse_json_file(json_fileos.path.abspath(sys.argv[1]))else:model_args, data_args, training_args parser.parse_args_into_dataclasses()if model_args.use_auth_token is not None:warnings.warn(The use_auth_token argument is deprecated and will be removed in v4.34. Please use token instead.,FutureWarning,)if model_args.token is not None:raise ValueError(token and use_auth_token are both specified. Please set only the argument token.)model_args.token model_args.use_auth_token# Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The# information sent is the one passed as arguments along with your Python/PyTorch versions.send_example_telemetry(run_mlm, model_args, data_args)# Setup logginglogging.basicConfig(format%(asctime)s - %(levelname)s - %(name)s - %(message)s,datefmt%m/%d/%Y %H:%M:%S,handlers[logging.StreamHandler(sys.stdout)],)if training_args.should_log:# The default of training_args.log_level is passive, so we set log level at info here to have that default.transformers.utils.logging.set_verbosity_info()log_level training_args.get_process_log_level()logger.setLevel(log_level)datasets.utils.logging.set_verbosity(log_level)transformers.utils.logging.set_verbosity(log_level)transformers.utils.logging.enable_default_handler()transformers.utils.logging.enable_explicit_format()# Log on each process the small summary:logger.warning(fProcess rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}, fdistributed training: {training_args.parallel_mode.value distributed}, 16-bits training: {training_args.fp16})# Set the verbosity to info of the Transformers logger (on main process only):logger.info(fTraining/evaluation parameters {training_args})# Detecting last checkpoint.last_checkpoint Noneif 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(fOutput 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(fCheckpoint 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 before initializing model.set_seed(training_args.seed)# Get the datasets: you can either provide your own CSV/JSON/TXT training and evaluation files (see below)# or just provide the name of one of the public datasets available on the hub at https://huggingface.co/datasets/# (the dataset will be downloaded automatically from the datasets Hub## For CSV/JSON files, this script will use the column called text or the first 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.dataset_name is not None:# Downloading and loading a dataset from the hub.raw_datasets load_dataset(data_args.dataset_name,data_args.dataset_config_name,cache_dirmodel_args.cache_dir,tokenmodel_args.token,streamingdata_args.streaming,)if validation not in raw_datasets.keys():raw_datasets[validation] load_dataset(data_args.dataset_name,data_args.dataset_config_name,splitftrain[:{data_args.validation_split_percentage}%],cache_dirmodel_args.cache_dir,tokenmodel_args.token,streamingdata_args.streaming,)raw_datasets[train] load_dataset(data_args.dataset_name,data_args.dataset_config_name,splitftrain[{data_args.validation_split_percentage}%:],cache_dirmodel_args.cache_dir,tokenmodel_args.token,streamingdata_args.streaming,)else:data_files {}if data_args.train_file is not None:data_files[train] data_args.train_fileextension data_args.train_file.split(.)[-1]if data_args.validation_file is not None:data_files[validation] data_args.validation_fileextension data_args.validation_file.split(.)[-1]if extension txt:extension textraw_datasets load_dataset(extension,data_filesdata_files,cache_dirmodel_args.cache_dir,tokenmodel_args.token,)# If no validation data is there, validation_split_percentage will be used to divide the dataset.if validation not in raw_datasets.keys():raw_datasets[validation] load_dataset(extension,data_filesdata_files,splitftrain[:{data_args.validation_split_percentage}%],cache_dirmodel_args.cache_dir,tokenmodel_args.token,)raw_datasets[train] load_dataset(extension,data_filesdata_files,splitftrain[{data_args.validation_split_percentage}%:],cache_dirmodel_args.cache_dir,tokenmodel_args.token,)# See more about loading any type of standard or custom dataset (from files, python dict, pandas DataFrame, etc) at# https://huggingface.co/docs/datasets/loading_datasets.# Load pretrained model and tokenizer## Distributed training:# The .from_pretrained methods guarantee that only one local process can concurrently# download model vocab.config_kwargs {cache_dir: model_args.cache_dir,revision: model_args.model_revision,token: model_args.token,trust_remote_code: model_args.trust_remote_code,}if model_args.config_name:config AutoConfig.from_pretrained(model_args.config_name, **config_kwargs)elif model_args.model_name_or_path:config AutoConfig.from_pretrained(model_args.model_name_or_path, **config_kwargs)else:config CONFIG_MAPPING[model_args.model_type]()logger.warning(You are instantiating a new config instance from scratch.)if model_args.config_overrides is not None:logger.info(fOverriding config: {model_args.config_overrides})config.update_from_string(model_args.config_overrides)logger.info(fNew config: {config})tokenizer_kwargs {cache_dir: model_args.cache_dir,use_fast: model_args.use_fast_tokenizer,revision: model_args.model_revision,token: model_args.token,trust_remote_code: model_args.trust_remote_code,}if model_args.tokenizer_name:tokenizer AutoTokenizer.from_pretrained(model_args.tokenizer_name, **tokenizer_kwargs)elif model_args.model_name_or_path:tokenizer AutoTokenizer.from_pretrained(model_args.model_name_or_path, **tokenizer_kwargs)else:raise ValueError(You are instantiating a new tokenizer from scratch. This is not supported by this script. You can do it from another script, save it, and load it from here, using --tokenizer_name.)if model_args.model_name_or_path:torch_dtype (model_args.torch_dtypeif model_args.torch_dtype in [auto, None]else getattr(torch, model_args.torch_dtype))model AutoModelForMaskedLM.from_pretrained(model_args.model_name_or_path,from_tfbool(.ckpt in model_args.model_name_or_path),configconfig,cache_dirmodel_args.cache_dir,revisionmodel_args.model_revision,tokenmodel_args.token,trust_remote_codemodel_args.trust_remote_code,torch_dtypetorch_dtype,low_cpu_mem_usagemodel_args.low_cpu_mem_usage,)else:logger.info(Training new model from scratch)model AutoModelForMaskedLM.from_config(config, trust_remote_codemodel_args.trust_remote_code)# We resize the embeddings only when necessary to avoid index errors. If you are creating a model from scratch# on a small vocab and want a smaller embedding size, remove this test.embedding_size model.get_input_embeddings().weight.shape[0]if len(tokenizer) embedding_size:model.resize_token_embeddings(len(tokenizer))# Preprocessing the datasets.# First we tokenize all the texts.if training_args.do_train:column_names list(raw_datasets[train].features)else:column_names list(raw_datasets[validation].features)text_column_name text if text in column_names else column_names[0]# text_column_name contentif data_args.max_seq_length is None:max_seq_length tokenizer.model_max_lengthif max_seq_length 1024:logger.warning(The chosen tokenizer supports a model_max_length that is longer than the default block_size value of 1024. If you would like to use a longer block_size up to tokenizer.model_max_length you can override this default with --block_size xxx.)max_seq_length 1024else:if data_args.max_seq_length tokenizer.model_max_length:logger.warning(fThe max_seq_length passed ({data_args.max_seq_length}) is larger than the maximum length for the fmodel ({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)if data_args.line_by_line:# When using line_by_line, we just tokenize each nonempty line.padding max_length if data_args.pad_to_max_length else Falsedef tokenize_function(examples):# Remove empty linesexamples[text_column_name] [line for line in examples[text_column_name] if len(line) 0 and not line.isspace()]return tokenizer(examples[text_column_name],paddingpadding,truncationTrue,max_lengthmax_seq_length,# We use this option because DataCollatorForLanguageModeling (see below) is more efficient when it# receives the special_tokens_mask.return_special_tokens_maskTrue,)with training_args.main_process_first(descdataset map tokenization):if not data_args.streaming:tokenized_datasets raw_datasets.map(tokenize_function,batchedTrue,num_procdata_args.preprocessing_num_workers,remove_columns[text_column_name],load_from_cache_filenot data_args.overwrite_cache,descRunning tokenizer on dataset line_by_line,)else:tokenized_datasets raw_datasets.map(tokenize_function,batchedTrue,remove_columns[text_column_name],)else:# Otherwise, we tokenize every text, then concatenate them together before splitting them in smaller parts.# We use return_special_tokens_maskTrue because DataCollatorForLanguageModeling (see below) is more# efficient when it receives the special_tokens_mask.def tokenize_function(examples):return tokenizer(examples[text_column_name], return_special_tokens_maskTrue)with training_args.main_process_first(descdataset map tokenization):if not data_args.streaming:tokenized_datasets raw_datasets.map(tokenize_function,batchedTrue,num_procdata_args.preprocessing_num_workers,remove_columnscolumn_names,load_from_cache_filenot data_args.overwrite_cache,descRunning tokenizer on every text in dataset,)else:tokenized_datasets raw_datasets.map(tokenize_function,batchedTrue,remove_columnscolumn_names,)# Main data processing function that will concatenate all texts from our dataset and generate chunks of# max_seq_length.def group_texts(examples):# Concatenate all texts.concatenated_examples {k: list(chain(*examples[k])) for k in examples.keys()}total_length len(concatenated_examples[list(examples.keys())[0]])# We drop the small remainder, and if the total_length max_seq_length we exclude this batch and return an empty dict.# We could add padding if the model supported it instead of this drop, you can customize this part to your needs.total_length (total_length // max_seq_length) * max_seq_length# Split by chunks of max_len.result {k: [t[i : i max_seq_length] for i in range(0, total_length, max_seq_length)]for k, t in concatenated_examples.items()}return result# Note that with batchedTrue, this map processes 1,000 texts together, so group_texts throws away a# remainder for each of those groups of 1,000 texts. You can adjust that batch_size here but a higher value# might be slower to preprocess.## To speed up this part, we use multiprocessing. See the documentation of the map method for more information:# https://huggingface.co/docs/datasets/process#mapwith training_args.main_process_first(descgrouping texts together):if not data_args.streaming:tokenized_datasets tokenized_datasets.map(group_texts,batchedTrue,num_procdata_args.preprocessing_num_workers,load_from_cache_filenot data_args.overwrite_cache,descfGrouping texts in chunks of {max_seq_length},)else:tokenized_datasets tokenized_datasets.map(group_texts,batchedTrue,)if training_args.do_train:if train not in tokenized_datasets:raise ValueError(--do_train requires a train dataset)train_dataset tokenized_datasets[train]if data_args.max_train_samples is not None:max_train_samples min(len(train_dataset), data_args.max_train_samples)train_dataset train_dataset.select(range(max_train_samples))if training_args.do_eval:if validation not in tokenized_datasets:raise ValueError(--do_eval requires a validation dataset)eval_dataset tokenized_datasets[validation]if data_args.max_eval_samples is not None:max_eval_samples min(len(eval_dataset), data_args.max_eval_samples)eval_dataset eval_dataset.select(range(max_eval_samples))def preprocess_logits_for_metrics(logits, labels):if isinstance(logits, tuple):# Depending on the model and config, logits may contain extra tensors,# like past_key_values, but logits always come firstlogits logits[0]return logits.argmax(dim-1)metric evaluate.load(accuracy, cache_dir./)def compute_metrics(eval_preds):preds, labels eval_preds# preds have the same shape as the labels, after the argmax(-1) has been calculated# by preprocess_logits_for_metricslabels labels.reshape(-1)preds preds.reshape(-1)mask labels ! -100labels labels[mask]preds preds[mask]return metric.compute(predictionspreds, referenceslabels)# Data collator# This one will take care of randomly masking the tokens.pad_to_multiple_of_8 data_args.line_by_line and training_args.fp16 and not data_args.pad_to_max_lengthdata_collator DataCollatorForLanguageModeling(tokenizertokenizer,mlm_probabilitydata_args.mlm_probability,pad_to_multiple_of8 if pad_to_multiple_of_8 else None,)# Initialize our Trainertrainer Trainer(modelmodel,argstraining_args,train_datasettrain_dataset if training_args.do_train else None,eval_dataseteval_dataset if training_args.do_eval else None,tokenizertokenizer,data_collatordata_collator,compute_metricscompute_metrics if training_args.do_eval and not is_torch_xla_available() else None,preprocess_logits_for_metricspreprocess_logits_for_metricsif training_args.do_eval and not is_torch_xla_available()else None,)# Trainingif training_args.do_train:checkpoint Noneif training_args.resume_from_checkpoint is not None:checkpoint training_args.resume_from_checkpointelif last_checkpoint is not None:checkpoint last_checkpointtrain_result trainer.train(resume_from_checkpointcheckpoint)trainer.save_model() # Saves the tokenizer too for easy uploadmetrics train_result.metricsmax_train_samples (data_args.max_train_samples if data_args.max_train_samples is not None else len(train_dataset))metrics[train_samples] min(max_train_samples, len(train_dataset))trainer.log_metrics(train, metrics)trainer.save_metrics(train, metrics)trainer.save_state()# Evaluationif training_args.do_eval:logger.info(*** Evaluate ***)metrics trainer.evaluate()max_eval_samples data_args.max_eval_samples if data_args.max_eval_samples is not None else len(eval_dataset)metrics[eval_samples] min(max_eval_samples, len(eval_dataset))try:perplexity math.exp(metrics[eval_loss])except OverflowError:perplexity float(inf)metrics[perplexity] perplexitytrainer.log_metrics(eval, metrics)trainer.save_metrics(eval, metrics)kwargs {finetuned_from: model_args.model_name_or_path, tasks: fill-mask}if data_args.dataset_name is not None:kwargs[dataset_tags] data_args.dataset_nameif data_args.dataset_config_name is not None:kwargs[dataset_args] data_args.dataset_config_namekwargs[dataset] f{data_args.dataset_name} {data_args.dataset_config_name}else:kwargs[dataset] data_args.dataset_nameif training_args.push_to_hub:trainer.push_to_hub(**kwargs)else:trainer.create_model_card(**kwargs)def _mp_fn(index):# For xla_spawn (TPUs)main()if __name__ __main__:main()
三. 训练 python run_mlm.py \--model_type bert \--tokenizer_name /home/chenjq/model/m3e-base \--train_file /home/chenjq/pythonWork/nlp/train_new_gpt2/2020-40_zh_head_0000.json \--num_train_epochs 2 \--per_device_train_batch_size 64 \--gradient_accumulation_steps 8 \--per_device_eval_batch_size 32 \--do_train \--save_steps 500 \--do_eval \--evaluation_strategy steps \--eval_steps 500\--weight_decay0.1 \--warmup_steps500 \--lr_scheduler_typecosine \--learning_rate 3e-3 \--logging_steps 100 \--fp16 \--max_seq_length 512 \--config_overrides hidden_size384,num_hidden_layers4,intermediate_size1024 \--output_dir ./output/test-clm-2
部分参数说明
--tokenizer_name 指定使用的的分词器本博客是用的是m3e模型的分词器需要提前下载好该模型并存储在对应目录
下载地址https://huggingface.co/moka-ai/m3e-base/tree/main --train_file 训练数据从第一部分的链接下载
--config_overrides hidden_size384,num_hidden_layers4,intermediate_size1024
这个参数是重点用于修改bert模型的参数
bert-base的hidden_size768num_hidden_layers12intermediate_size3076
我们的目的是为了训练一个tiny-bert通过对这几个参数的修改我们可以获得一个迷你版的bert 本实验我大概使用了2G的数据训练了2个epoch
