Bert Base Cased Trec Coarse

🚀 bert-base-cased在TREC 6类任务上的训练模型

本项目是一个在“trec”数据集上训练的简单基础BERT模型，可用于文本分类任务，在测试集上取得了较高的准确率。

🚀 快速开始

模型描述

这是一个在“trec”数据集上训练的简单基础BERT模型。

预期用途和限制

使用方法

Transformers

# 加载模型和分词器
from transformers import AutoModelForSequenceClassification, AutoTokenizer

model = AutoModelForQuestionAnswering.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

# 使用管道
from transformers import pipeline

model_name = "aychang/bert-base-cased-trec-coarse"

nlp = pipeline("sentiment-analysis", model=model_name, tokenizer=model_name)

results = nlp(["Where did the queen go?", "Why did the Queen hire 1000 ML Engineers?"])

AdaptNLP

from adaptnlp import EasySequenceClassifier

model_name = "aychang/bert-base-cased-trec-coarse"
texts = ["Where did the queen go?", "Why did the Queen hire 1000 ML Engineers?"]

classifer = EasySequenceClassifier
results = classifier.tag_text(text=texts, model_name_or_path=model_name, mini_batch_size=2)

限制和偏差

这是一个在基准数据集上训练的最小语言模型。

训练数据

TREC https://huggingface.co/datasets/trec

训练过程

包括预处理、使用的硬件、超参数等。

硬件

使用了一块V100。

超参数和训练参数

from transformers import TrainingArguments

training_args = TrainingArguments(
    output_dir='./models',
    num_train_epochs=2,
    per_device_train_batch_size=16,
    per_device_eval_batch_size=16,
    warmup_steps=500,
    weight_decay=0.01,
    evaluation_strategy="steps",
    logging_dir='./logs',
    save_steps=3000
)

评估结果

{'epoch': 2.0,
 'eval_accuracy': 0.974,
 'eval_f1': array([0.98181818, 0.94444444, 1.        , 0.99236641, 0.96995708,
        0.98159509]),
 'eval_loss': 0.138086199760437,
 'eval_precision': array([0.98540146, 0.98837209, 1.        , 0.98484848, 0.94166667,
        0.97560976]),
 'eval_recall': array([0.97826087, 0.90425532, 1.        , 1.        , 1.        ,
        0.98765432]),
 'eval_runtime': 1.6132,
 'eval_samples_per_second': 309.943}

📄 许可证

本项目采用MIT许可证。

📋 模型信息