'BertQA' object has no attribute 'adam_epsilon'

[nehabharambe]

I am trying to train the CDQA with custom json file - downloaded from cdqa annotator and I am getting above mentioned error at the last line of code,please find the code as follows:

`from cdqa.pipeline.cdqa_sklearn import QAPipeline from cdqa.utils.download import download_model

download_model(model='bert-squad_1.1', dir='./models')

cdqa_pipeline = QAPipeline(reader='/content/gdrive/My Drive/BERT_Training/models/bert_qa.joblib')

path_to_data = '/content/cdqa-v1.1-2.json' cdqa_pipeline.fit_reader(path_to_data)`

[andrelmfarias]

Hi,

It was a problem with the available models. I have uploaded a fixed version.

[nehabharambe]

Hi,

I am getting a different runtime error, can you help me with this error if I am using GPU for training the new model - bert_qa.joblib. (I am using Google colab for runnning my models)

RuntimeError: Expected object of backend CUDA but got backend CPU for argument #3 'index'

If I change my runtime and switch to CPU, the system keeps crashings while excuting the code, is it possible to run this code on GPU?

I found a similar issue on github - https://github.com/huggingface/transformers/issues/227 (refer this link), but couldn't resolve this problem.

[andrelmfarias]

I have just fixed the issue, I think you can try it again.

[ivopisarovic]

Thanks, you saved me! ❤️