Pure White
Pure White
https://github.com/alibaba/EasyNLP/blob/bf17be30b01ec3ebf7a135804b4fb0fff544cb7f/easynlp/fewshot_learning/fewshot_evaluator.py#L131-L145 It's due to the appearance of another argument called `user_defined_parameters` in `kwargs`, IMO. Add a line `kwargs.pop('user_defined_parameters')` before generating `FewshotBaseDataset` may fix it.
Hi, I'm tring to do further pretraining with XLNet too. I had done the same thing with BERT before and confirm its effectiveness, so I wonder why you got worse...
> 升级了torch版本后可以了。但项目主页写的,torch==1.7,transformer==4.26.1 你好,请问你升级到多少版本以后正常