allen
allen
Agree with @rahul-iisc at test time decoder_inputs is the same as encoder inputs running main.py we have test precision about 63% (Step: 99900) At training time we have two options:...
Might be some mistakes? Even using huggingface basic MLM pipline continue pretrain on deberta-v3-large works. Tested on kaggle nbme dataset.
for those willing to find the code, you can refer to this nice work https://github.com/abisee/pointer-generator I will implement new pointer-generator related work based on tf 1.2 attention_wrapper, that is dynamic...