
Reputation
Badges 1
Eureka!@<1523701205467926528:profile|AgitatedDove14> Would this be equivalent to an automated job submission from clearml to the cluster? My cluster security rules do not allow for automated job submission. I am looking for a setup which allows me to essentially create the workers and start the tasks from a slurm script, with clearml simply receiving the information about the workers and sending information to the cluster regarding allotment of the tasks, but without clearml explicitly sending the ...
if anyone is curious, I figured out a solution in which I can simply define hyperparameters that I might not actually use depending on the value of num_layers. This works because of the limited size range, but curious to think about this more for higher dynamic ranges.