-
Notifications
You must be signed in to change notification settings - Fork 70
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
about DecAtt #29
Comments
DecAtt is very difficult to train, which I tried many ways to make it work, including gradient clipping, sorted length and etc. Previously people used length sorting to accelerate the model training and convergence speed, since the input doesn't vary a lot. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
When i use this model for wikiQA Task,i found that the batch list is difficult.
![image](https://user-images.githubusercontent.com/20167430/61512696-4dfa2480-aa2d-11e9-9e59-6d5e14eebb98.png)
![image](https://user-images.githubusercontent.com/20167430/61512742-708c3d80-aa2d-11e9-8e86-53e1a97e464d.png)
Why should we resort the length?And The interval of batch_list is not 32.
The text was updated successfully, but these errors were encountered: