AttributeError: 'ATEPCConfigManager' object has no attribute 'hidden_dim'

[sqs17]

AttributeError: 'ATEPCConfigManager' object has no attribute 'hidden_dim'

[sqs17]

and when i add "config.hidden_dim = 768"
error is :
Traceback (most recent call last):
File "c:\PyABSA-2\yelp\test.py", line 57, in
trainer = ATEPC.ATEPCTrainer(
File "D:\anaconda\envs\pyabsa2\lib\site-packages\pyabsa\tasks\AspectTermExtraction\trainer\atepc_trainer.py", line 69, in init
self._run()
File "D:\anaconda\envs\pyabsa2\lib\site-packages\pyabsa\framework\trainer_class\trainer_template.py", line 240, in _run
model_path.append(self.training_instructor(self.config).run())
File "D:\anaconda\envs\pyabsa2\lib\site-packages\pyabsa\tasks\AspectTermExtraction\instructor\atepc_instructor.py", line 799, in run
return self._train(criterion=None)
File "D:\anaconda\envs\pyabsa2\lib\site-packages\pyabsa\framework\instructor_class\instructor_template.py", line 353, in _train
self._resume_from_checkpoint()
File "D:\anaconda\envs\pyabsa2\lib\site-packages\pyabsa\framework\instructor_class\instructor_template.py", line 451, in _resume_from_checkpoint
self.model.load_state_dict(
File "D:\anaconda\envs\pyabsa2\lib\site-packages\torch\nn\modules\module.py", line 2153, in load_state_dict
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for FAST_LCF_ATEPC:
size mismatch for dense.weight: copying a param with shape torch.Size([3, 768]) from checkpoint, the shape in current model is torch.Size([2, 768]).
size mismatch for dense.bias: copying a param with shape torch.Size([3]) from checkpoint, the shape in current model is torch.Size([2]).

[sqs17]

I downloaded the model locally, using microsoft\deberta-v3-base

[sqs17]

Its my checkpoints.

[sqs17]

This seems to have something to do with the training data set. I successfully fine-tuned the training data when I used 119.Yelp and 133.finNews, but I reported errors when I used my own data set and 99.PoliticalData. 99.PoliticalData has dimension 4, and my dataset has dimension 1

[yangheng95]

Please make sure that your datasets contains the same number of the labels as the pretrained model you are going to use. If the number of labels are not equal, please try fine-tuning the model based on merely your own dataset.

[sqs17]

Hello, thank you very much for your answer.
The task I want to do now is to extract the aspect level of each sentence and conduct sentiment analysis. A sentence corresponds to multiple aspect levels. I don't quite understand the meaning of the number of labels, doesn't every sentence correspond to one aspect? If it corresponds to multiple aspects, repeat the sentence. Thanks again for your help.

[sqs17]

是指这里的output_dim吗

[sqs17]

Thanks for your suggestion, after modifying self.output_dim, my code runs well. Is this because the result of the pre-trained model is a triplet of this type? But my data is consistent with the data structure of the pre-trained model.