Restore from s3 bucket without resume

Hello everyone, I’m trying to do incremental training (more iterations) using a previously trained model’s checkpoint. The checkpoint is stored in an s3 bucket, and I would like to restore it without resume to avoid instant termination as suggested here. When I try to do that, I’m getting the following error: FileNotFoundError: [Errno Path does not exist]

Is it a bug? There is a way to do that?

Hmm, this does seem like a bug (sorry for the slow reply).

Could you file an issue on github?