No sapace left on colab VM

[ANTONIOPSD]

It happens when I save too many steps, there should be an option to delete old checkpoints from the colab disk once they are uploaded to gdrive otherwise the colab disk gets filled and the process fails.

This time happened at the end of the training because I set the backup checkpoints specifically to fill the disk and it failed at generating the tokenizer:

My gdrive has more than 60GB left

The final ckpt is created in gdrive and the tokenizer folder is empty