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 have seen the explanation: "inner states need to be record, type: tuple". Are there any other role except for recording the innner states? Because I found it can still get the correct result without the parameter. looking forword to your reply. THX.
The text was updated successfully, but these errors were encountered:
The structure "inner_packed_states_array" is an interface for future expand.
Now, for Transformer model, this structure could be ignored because every token records full history labels. So, you could get correct results without the parameters.
However for other End-to-End models, for example LAS, it is useful to record some Intermediate state to avoid Repeated calculation
I have seen the explanation: "inner states need to be record, type: tuple". Are there any other role except for recording the innner states? Because I found it can still get the correct result without the parameter. looking forword to your reply. THX.
The text was updated successfully, but these errors were encountered: