search

macournoyer / neuralconvo

Neural conversational model in Torch
776 stars 347 forks source link

Resume training from a saved model #55

Closed nabihach closed 8 years ago

nabihach commented 8 years ago

Is there a way to resume training from where it was previously left off? For example, I trained a particular model for 50 epochs and saved it to model.t7. I tested it using eval.lua and discovered that it needed training for another 50 epochs. Is there an easy way for me to resume training such that the model is initialized with model.t7 and then continues to train?

macournoyer commented 8 years ago

Not at the moment, but it should be possible to implement relatively easily in train.lua:

Replace the line that initializes the model w/ code that loads it from the .t7 file (like in eval.lua).

But the epoch count will count back from 1...

nabihach commented 8 years ago

Thanks. It turns out that it is also possible to resume training on a different dataset this way.