无法从“变压器”导入“AutoModelForSequenceClassification”

Posted 2023-03-04

【中文标题】无法从“变压器”导入“AutoModelForSequenceClassification”

【英文标题】：cannot import 'AutoModelForSequenceClassification' from 'transformers'

【发布时间】：2021-06-28 18:17:30

【问题描述】：

代码是

from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline

t = AutoTokenizer.from_pretrained('/some/directory')
m = AutoModelForSequenceClassification.from_pretrained('/some/directory')
c2 = pipeline(task = 'sentiment-analysis', model=m, tokenizer=t)

错误是

cannot import 'AutoModelForSequenceClassification' from 'transformers'

【问题讨论】：

什么是变形金刚版本？

它正在工作。我已经安装了 4.3.3

【参考方案1】：

from transformers import AutoModelForSequenceClassification, BertForSequenceClassification
from transformers import (XLMRobertaConfig, XLMRobertaTokenizer, TFXLMRobertaModel)            
from transformers import AutoTokenizer, AutoConfig, TFAutoModel    

PRETRAINED_MODEL_TYPES = 
    'xlmroberta': (AutoConfig, AutoModelForSequenceClassification, AutoTokenizer, 'akhooli/xlm-r-large-arabic-toxic')


# model_class,model = AutoModelForSequenceClassification.from_pretrained("akhooli/xlm-r-large-arabic-toxic")

config_class, model_class, tokenizer_class, model_name = PRETRAINED_MODEL_TYPES['xlmroberta']

# Download vocabulary from huggingface.co and cache.
tokenizer = AutoTokenizer.from_pretrained(model_name,use_fast=False) #fast tokenizer

tokenizer