Dear users,

this is a reminder that today, Friday October 30th, from 15:00 to 19:00, the University will carry out an extraordinary maintenance on the firewalls of the University network. During this window the whole University network will be unavailable.

The internal network of the cluster is not affected. Running and queued Slurm jobs will keep going, and access to storage and communication between nodes, including multi-node training, will work as usual.

Jobs that need connections outside the cluster, on the other hand, may run into problems or fail. If your jobs run during this window, we recommend downloading the resources you need in advance, setting HF_HUB_OFFLINE=1 for models already in your cache, and switching to offline logging (WANDB_MODE=offline, then wandb sync once the maintenance is over).

Please also note that, since the University network will be down, connecting to the login nodes from outside may not be possible during the maintenance.

Best regards,
Lorenzo.