i follow the chain of the code decode_with_lm
,and want to know how the language model used to update the model weight.But i only see the code optimizer.compute_gradients(avg_loss)
,in other words,it seems that loss used to update the model weight.so my question is whether is LM used to update model parameters per epoch? which one is right below?
-
LM that base on the output of the network is used to train model and update the model parameter?
-
LM change the output of the network and then show the most probable result to us according to the statistical text corpus?
if i want to add LM to my own model ,how can i do it?
Thank you in advance!!!