Continuing training from a wrapped checkpoint does not work.

[Taikakim]

I'm getting a RuntimeError: PytorchStreamReader failed reading zip archive: failed finding central directory error when trying to resume from a wrapped checkpoint. The only values I changed in the config are LR and inv_gamma.

The unwrap script is complaining about : RuntimeError: PytorchStreamReader failed reading zip archive: failed finding central directory

The checkpoints are on my Drive and I'm running in Colab, but I don't recall this was a problem before.

[Taikakim]

Ah, problem solved, the upload from Drive was interrupted before the checkpoint had been transferred :/ (I saved the checkpoint from inside a cell, and apparently, even if it shows in the right place in Colab, unless it's synced and I disconnect the VM, Colab does not finish uploads in the background.)