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
As you can see in the highlighted code, the parameter "bias" is always searched in the config, but in my experience the bias doesn't exist for the output layer more often than yes. Also, when the parameter name is not found, the program just crashes with a criticalError.
As a side note, the error message doesn't show the parameter name causing the error.
The text was updated successfully, but these errors were encountered:
As mentioned in #12, there is other code assuming the existence of this tensor. Should we create a zero bias tensor when it doesn't exist, or more cleanly check the existence of this tensor? In the second case, the retrieval of the bias tensor should be able to handle the error.
Also making another class for the case without a bias is an easy option to follow.
rasr/src/Lm/BlasNceSoftmaxAdapter.cc
Line 25 in 473d202
As you can see in the highlighted code, the parameter "bias" is always searched in the config, but in my experience the bias doesn't exist for the output layer more often than yes. Also, when the parameter name is not found, the program just crashes with a criticalError.
As a side note, the error message doesn't show the parameter name causing the error.
The text was updated successfully, but these errors were encountered: