Finetuning script at the Kaggle notebook provides no output

There is error in the Kaggle notebook with example code for "Emojich" finetuning, that published here:
https://www.kaggle.com/shonenkov/emojich-rudall-e
When finetuning with a custom dataset, on the first try the training was interrupted early, because of the line:
`EARLY_STOP = True`
This value was changed to "False", but then training continues until the kernel stops after 6 hours and 34 minutes, due to limitation in Kaggle. After that, there is no output data available.
The question is: how to set training to stop after a certain time (or a certain number of steps or epochs) to make saved checkpoints available after the end of execution?