Hi!
How to correctly configure Clearml with PyTorch-Ignite to write checkpoints to the /mnt/nfs/shared Project Dir in a 3-agent cluster?
I tried this

task = Task.init(
        project_name="train-unet-giraffe",
        task_name=f"split-{config.data.split_index}",
        output_uri=config.clearml.output_uri,
    )
...
    last_checkpointer = Checkpoint(
        to_save={"model": model, "optimizer": optimizer, "trainer": trainer},
        save_handler=ClearMLSaver(create_dir=True),
        n_saved=2,
        filename_prefix="last",
        global_step_transform=global_step_from_engine(trainer),
    )

But the ClearMLSaver uses a temporary folder per agent which is hard to collect afterwards.