Sorry to bother, I met this error when I evaluate some models, and I didn't find a good method to fix it.
What does 'MaskedLMOutput'means?Could somebody tell me How to fix this please? Thank you.
(AttributeError: 'MaskedLMOutput' object has no attribute 'view')
from transformers import BertForMaskedLM
class BertPunc(nn.Module):
def __init__(self, segment_size, output_size, dropout):
super(BertPunc, self).__init__()
self.bert = BertForMaskedLM.from_pretrained('cl-tohoku/bert-base-japanese')
self.bert_vocab_size = 32000
self.bn = nn.BatchNorm1d(segment_size*self.bert_vocab_size)
self.fc = nn.Linear(segment_size*self.bert_vocab_size, output_size)
self.dropout = nn.Dropout(dropout)
def forward(self, input):
x = self.bert(input)
x = x.view(x.shape[0], -1) # wrong thing here
x = self.fc(self.dropout(self.bn(x)))
return x
I ran this in jupyter notebook:
def predictions(data_loader):
y_pred = []
y_true = []
for inputs, labels in tqdm(data_loader, total=len(data_loader)):
with torch.no_grad():
inputs, labels = inputs.cuda(), labels.cuda()
output = bert_punc(inputs)
y_pred += list(output.argmax(dim=1).cpu().data.numpy().flatten())
y_true += list(labels.cpu().data.numpy().flatten())
return y_pred, y_true
def evaluation(y_pred, y_test):
precision, recall, f1, _ = metrics.precision_recall_fscore_support(
y_test, y_pred, average=None, labels=[1, 2, 3])
overall = metrics.precision_recall_fscore_support(
y_test, y_pred, average='macro', labels=[1, 2, 3])
result = pd.DataFrame(
np.array([precision, recall, f1]),
columns=list(punctuation_enc.keys())[1:],
index=['Precision', 'Recall', 'F1']
)
result['OVERALL'] = overall[:3]
return result
y_pred_test, y_true_test = predictions(data_loader_test)
eval_test = evaluation(y_pred_test, y_true_test)
eval_test
wrong here:
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
Input In [12], in <cell line: 1>()
----> 1 y_pred_test, y_true_test = predictions(data_loader_test)
2 eval_test = evaluation(y_pred_test, y_true_test)
3 eval_test
Input In [10], in predictions(data_loader)
5 with torch.no_grad():
6 inputs, labels = inputs.cuda(), labels.cuda()
----> 7 output = bert_punc(inputs)
8 y_pred += list(output.argmax(dim=1).cpu().data.numpy().flatten())
9 y_true += list(labels.cpu().data.numpy().flatten())
File ~\anaconda3\lib\site-packages\torch\nn\modules\module.py:1110, in Module._call_impl(self, *input, **kwargs)
1106 # If we don't have any hooks, we want to skip the rest of the logic in
1107 # this function, and just call forward.
1108 if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
1109 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1110 return forward_call(*input, **kwargs)
1111 # Do not call functions when jit is used
1112 full_backward_hooks, non_full_backward_hooks = [], []
File ~\anaconda3\lib\site-packages\torch\nn\parallel\data_parallel.py:166, in DataParallel.forward(self, *inputs, **kwargs)
163 kwargs = ({},)
165 if len(self.device_ids) == 1:
--> 166 return self.module(*inputs[0], **kwargs[0])
167 replicas = self.replicate(self.module, self.device_ids[:len(inputs)])
168 outputs = self.parallel_apply(replicas, inputs, kwargs)
File ~\anaconda3\lib\site-packages\torch\nn\modules\module.py:1110, in Module._call_impl(self, *input, **kwargs)
1106 # If we don't have any hooks, we want to skip the rest of the logic in
1107 # this function, and just call forward.
1108 if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
1109 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1110 return forward_call(*input, **kwargs)
1111 # Do not call functions when jit is used
1112 full_backward_hooks, non_full_backward_hooks = [], []
File F:\BertPunc-master\model.py:19, in BertPunc.forward(self, input)
16 def forward(self, input):
18 x = self.bert(input)
---> 19 x = x.view(x.shape[0], -1)
20 x = self.fc(self.dropout(self.bn(x)))
21 return x
AttributeError: 'MaskedLMOutput' object has no attribute 'view'
You can refer to the documentation of MaskedLMOutput. Basically, it is an object holding the loss, logits, hidden_states and attentions. It is not a tensor so you are getting this error. I think you are interested in logits, i.e., score for each token before applying softmax. Then, in the forward function, you can just access to the logits tensor like this
x = self.bert(input).logits