about val score
cikers opened this issue · 5 comments
Thank you very much for sharing your work!
I have a question.
I train the model in train_mode='train' and train_split='train', other settings are not change.
but the best score is no more than 65% in the val dataset.
In your work, the score is 66~67% in the val dataset.
What should i do for my confusion?
Thanks again!
can you provide the log file results/log/log_run_<VERSION>.txt for your training?
log_run_30.txt
I use resume=True to train the model, so the first val score is up to 60.21%.
It seems that you have modified the learning rate and removed the warm-up strategy for learning rate, which are very important to the final performance. I suggest checkout the latest version and use the default hyper-param instead to rerun the script
ok, I will check the code again.
Thank you.
Have you reproduced the results with the default hyper-params?