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 encountered a problem during the process of reproducing this code. In the second training stage, the output of the old model for new data was nan. I debugged the code and found that the distribution of the model's output data was significantly different, ultimately leading to numerical overflow. May I ask what tricks were used in your implementation process to avoid such issues.
The text was updated successfully, but these errors were encountered:
I encountered a problem during the process of reproducing this code. In the second training stage, the output of the old model for new data was nan. I debugged the code and found that the distribution of the model's output data was significantly different, ultimately leading to numerical overflow. May I ask what tricks were used in your implementation process to avoid such issues.
The text was updated successfully, but these errors were encountered: