Checkpoint set and restore
args.restore_step is referenced in train.py and model.py. Is there any documentation on recommended use for this argument in checkpointing? Thanks!
@rspiewak47 It is for the case you want to restart the training from an available checkpoint.
args.restore_step is referenced in train.py and model.py. Is there any documentation on recommended use for this argument in checkpointing? Thanks!
it will save checkpoint automatically each 100000 steps, you can change save_step in /config/LJSpeech/train.yaml to make it save earlier or later
to restore and continue training, put the checkpoint file (ex: 15000.pth.tar) in /output/ckpt/LJSpeech and add --restore_step 15000 at the end of train command