You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I am training a BERT-base model for Chinese. Default MLM and NSP tasks are used. I am trying to train the model for 96k steps to see if it benefits from longer training procedure. However, from step 65600 to step 65700, the MLM accuracy drops dramatically from 0.827 to 0.774 while the NSP accuracy remains high and stable. I am wondering how the drop takes place.
I have around 226k sentences in the original corpus and each one is split into two parts from the middle, just like book_review_bert.txt. During data preprocess, I modified the dup_factor from 5 to 50 to ensure diversity. The actual batch_size is [16 (args.batch_size) x 2 (args.world_size) x 1 (args.accumulation_steps)].
The text was updated successfully, but these errors were encountered:
Maybe the NSP accuracy remains high and stable because it is an easy task compared to MLM.
There is still a sudden rise of NSP loss, increasing from 0.010 (and less) to 0.017 in step 65700. MLM loss also increases remarkably, from 0.7 to 1.0. I think it is safe to say that both MLM and NSP suffer from a degradation. The only reason I come up with is that an error in a small fraction of the corpus occurs.
I am training a BERT-base model for Chinese. Default MLM and NSP tasks are used. I am trying to train the model for 96k steps to see if it benefits from longer training procedure. However, from step 65600 to step 65700, the MLM accuracy drops dramatically from 0.827 to 0.774 while the NSP accuracy remains high and stable. I am wondering how the drop takes place.
I have around 226k sentences in the original corpus and each one is split into two parts from the middle, just like book_review_bert.txt. During data preprocess, I modified the dup_factor from 5 to 50 to ensure diversity. The actual batch_size is [16 (args.batch_size) x 2 (args.world_size) x 1 (args.accumulation_steps)].
The text was updated successfully, but these errors were encountered: