abisee
wrong for increasing hidden_dim,emb_dim
Sorry for bothering you. In the early training, I set the 'hidden_dim' as 64, and set the 'emb_dim' as 32. Then ,when i increase them as '256' and '128'. it has an error. It seems that the shape of parameter generated by training in the early training is not fit for the later training. How can i deal with it.
You can't (easily) increase those things. You've been training a model that is performing matrix transformations on hidden vectors of size 64. You can't use that model to handle hidden vectors of size 256. You'd need different shaped weight matrices for that.
Perhaps you're confused because you can increase max number of encoder steps and max number of decoder steps. That's because increasing those things just means unrolling the RNN for more steps.
