大模型微调系列 --关于transformer中的pipeline是什么

• pipeline就是一个帮你做数据预处理以及一些模型加载，模型预测的步骤的工具，提供各种接口

定义数据预处理函数

• 对输入的文本进行预处理
  • 转换为小写
  • 去除标点符号
  • 去除特殊字符

def preprocess_text(text:str):text=text.lower()text=''.join([c for c in text if c.isalnum() or c.isspace()])  return text

模型加载的函数

• 利用transformer库加载一个模型

from transformers import AutoModelForSequenceClassification,AutoTokenizer  
def load_model_tokenizer(model_name):tokenizer = AutoTokenizer.from_pretrained(model_name)model = AutoModelForSequenceClassification.from_pretrained(model_name)return tokenizer, model

模型的预测

• 使用模型对预处理的文本进行预测

import torch 
def predict(text,model,tokenizer):text=preprocess_text(text)  inputs=tokenizer(text,return_tensors="pt")  outputs=model(**inputs)  _,predicted_class=torch.max(outputs.logits.softmax(dim=-1),dim=1)  return predicted_class  

组装

def self_pipe_line(text,model_name='./model/gpt2'):tokenizer,model=load_model_tokenizer(model_name)  return predict(text,model,tokenizer)  

使用的例子

text_to_classify="i am your father"  
predicted_class=self_pipe_line(text_to_classify)  
print(f"predict class: {predicted_class}")

Some weights of GPT2ForSequenceClassification were not initialized from the model checkpoint at ./model/gpt2 and are newly initialized: ['score.weight']
You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.predict class: tensor([1])