continue training from checkpoint?

[Zoeyyao27]

Is there a way to continue training from the last model

[vdobrovolskii]

To actually continue, you would need the saved optimizer states, which were removed to make the weights file smaller. But it should be possible to initialize the weights of the model using the pretrained weights and then start the training

[Zoeyyao27]

To actually continue, you would need the saved optimizer states, which were removed to make the weights file smaller. But it should be possible to initialize the weights of the model using the pretrained weights and then start the training

Thank you！ I actually managed to continue from saved checkpoint. But according to your code in https://github.com/vdobrovolskii/wl-coref/blob/96b3d5e50367b253cafa81ea82ae39e44ca7fa46/coref/coref_model.py#L266 I assume the optimizer states have already been saved?

[vdobrovolskii]

They were originally, but then they were stripped from the state dict by a separate script to save space

[Zoeyyao27]

Ha! I see. Thank you for your kind reply! Great work by the way! :)