Open S1mplePike opened 2 years ago
Hi Pike,
Thanks for your interest! I just found that the number of epochs to train should be 50. In the script I made that 5 instead. Now I corrected it. All other hyperparameters are the same as what I used in my experiment. I don't think the version of transformers makes any difference.
One reason I can think of is the number of GPUs to use. Were you using 4 GPUs as I did? Single-GPU training results should be different from multi-GPU training.
Hi, author. Recently I am follow this work. but,I got some problem. I cann't reproduction the results of VAST dataset f1 score=0.736 and f0 score=0.753, I wonder if the problem is caused by the version. my transformers package version == 4.18.0. I' d like to ask about the python packages version of experimental environment. and the hyperparameters of experiment.