How can I continue training from pretrained model?

[erogol]

Does the library provide a function to continue the training from a pretrained model?

[mlajtos]

Yes. Just use -init_from <path_to_checkpoint> train option.

https://github.com/jcjohnson/torch-rnn/blob/master/train.lua#L20

[dgcrouse]

That currently does not re-initialize all parameters, you need to look at what is saved and what is not. I am working on a new checkpointing system that will make resuming much easier, so stay tuned.