About training epochs.
hufengshuo07 opened this issue · 7 comments
I found that there isn't well-trained checkpoints, so I need train from scratch, could you tell me total number of iterations or epochs to approach your 84% results?
Hi, it converges after about 50 epochs (learning rate drops two times).
@Daniil-Osokin Oh, thanks. For myself, my training takes 38 hours and 70epochs on 1xP100 to converge with batchsize=64, pckh~84.4. Hope to help others
Thanks for the report back your results!
@Daniil-Osokin Oh, thanks. For myself, my training takes 38 hours and 70epochs on 1xP100 to converge with batchsize=64, pckh~84.4. Hope to help others
I have the same probelm for loading pre_trained model for initialization.
could you provide your pre_trained model for us? thank you so much
is it possible upload the model to Google Drive or Baidu Drive?
I found that there isn't well-trained checkpoints, so I need train from scratch, could you tell me total number of iterations or epochs to approach your 84% results?
I have the same probelm for loading pre_trained model for initialization.
could you provide your pre_trained model for us? thank you so much
is it possible upload the model to Google Drive or Baidu Drive?
Hi, @qintao361! I will try to do it next week.
Check it out: #12, 70 epochs and 84.57 PCKh.