【3】使用pytorch_lightning+transformers+torchmetric+datasets进行文本分类

183****0229

transformers中AutoModelForSequenceClassification的使用

# 导包
import torch
from transformers import AutoModelForSequenceClassification,AutoTokenizer

# huggingface.co上的模型
pretrained_name_or_path = "junnyu/roformer_chinese_base"

# 加载预训练的tokenizer
tokenizer = AutoTokenizer.from_pretrained(pretrained_name_or_path)

# 加载预训练的model,并定义进行15分类
model = AutoModelForSequenceClassification.from_pretrained(pretrained_name_or_path, num_labels=15)

# 输入的文本
text_list = ["江疏影甜甜圈自拍，迷之角度竟这么好看，美吸引一切事物","军嫂探亲拧包入住，部队家属临时来队房标准有了规定，全面落实！"]
labels = torch.tensor([1,2])
# 进行tokenizer
tokenized_inputs = tokenizer(text_list,padding=True,return_tensors="pt")
print(tokenized_inputs)

with torch.no_grad():
    outputs = model(**tokenized_inputs,labels=labels)
print(outputs.loss)
print(outputs.logits.shape) # shape [2,15]