【Foundation】（一）transformers之pipeline

1、介绍

本篇博客内容以及后续内容均来自b站up主你可是处女座啊

什么是pipeline?

• 将数据预处理、模型调用，结果后处理三部分组装成的流水线

• 是我们能够直接输入文本就能够获得最终答案

 非常类似于torch.nn中的sequencial

能够做什么？

支持任务类型：

• text-classification
• token-classification
• question-answering
• fill-mask
• translation
• text2text-generation
• text-generation
• conversatinal
• table-question-answering
• zero-shot-classification

2、代码实战

2.1、查看pipeline支持的任务类型

from transformers.pipelines import SUPPORTED_TASKS
for k,v in SUPPORTED_TASKS.items():
    print(k,v['type'])

2.2、pipeline的创建与使用方式

from transformers import pipeline
#根据任务类型直接创建pipeline，默认都是英文模型
pipe = pipeline("text-classification")
pipe('very sad')

2.3、指定任务类型，指定模型创建pipeline

 这里需要科学上网

pipe = pipeline("text-classification",model='uer/roberta-base-finetuned-dianping-chinese')
pipe('今天天气真好')

2.4、预先加载模型，在创建pipeline

from transformers import AutoModelForSequenceClassification,AutoTokenizer,pipeline

model = AutoModelForSequenceClassification.from_pretrained('uer/roberta-base-finetuned-dianping-chinese')
tokenizer = AutoTokenizer.from_pretrained('uer/roberta-base-finetuned-dianping-chinese')

pipe = pipeline('text-classification',model=model,tokenizer=tokenizer,device=0)

进行测试

pipe('我觉得不太行')

2.5、背后实现原理

2.5.1、导包

from transformers import AutoModelForSequenceClassification,AutoTokenizer,pipeline

model = AutoModelForSequenceClassification.from_pretrained('uer/roberta-base-finetuned-dianping-chinese')
tokenizer = AutoTokenizer.from_pretrained('uer/roberta-base-finetuned-dianping-chinese')

pipe = pipeline('text-classification',model=model,tokenizer=tokenizer)

2.5.2、对输入进行处理

input_text = "我觉得不太行"
inputs = tokenizer(input_text,return_tensors='pt')
inputs

2.5.3、获得模型输出

res = model(**inputs)
res

2.5.4、计算逻辑得分

logits = res.logits

logits = torch.softmax(logits,dim=-1)
logits

2.5.4、获得预测结果

pred = torch.argmax(logits).item()
pred

2.5.5、转化为真实标签

model.config.id2label
result = model.config.id2label.get(pred)
result