import torch | |
from transformers import MLukeTokenizer | |
from torch import nn | |
tokenizer = MLukeTokenizer.from_pretrained('studio-ousia/luke-japanese-base-lite') | |
model = torch.load('C:\\[modelのあるディレクトリ]\\My_luke_model_pn.pth') | |
text=input() | |
encoded_dict = tokenizer.encode_plus( | |
text, | |
return_attention_mask = True, # Attention maksの作成 | |
return_tensors = 'pt', # Pytorch tensorsで返す | |
) | |
pre = model(encoded_dict['input_ids'], token_type_ids=None, attention_mask=encoded_dict['attention_mask']) | |
SOFTMAX=nn.Softmax(dim=0) | |
num=SOFTMAX(pre.logits[0]) | |
if num[1]>0.5: | |
print(str(num[1])) | |
print('ポジティブ') | |
else: | |
print(str(num[1])) | |
print('ネガティブ') |