Hi,
I have trained my Transformers model until 140.000 steps. So now my saved models are from steps 138.398, 138.627, 139.008, 139.319, 139.550, 139.781, and 140.000. Is there any way where I can do “onmt-main infer” from step 110.000? Because after I check the tensorboard, my model’s best performance was on step 110.000. Thank you
Hi, look for the command line option --checkpoint_path.
Yes, I understand we can do that.
But in my case, I simply could not because the file is of mode.ckpt-110000 is not exist anymore.
Here is my checpoint file
Ah right so there is no way to use checkpoints that don’t exist. Next training you can increase the keep_checkpoint_max parameter in the train section of the configuration.