Running distributed tensorflow on machines with different versions of Tensorflow -
in particular i'm interested in happens if tensorflow installed on parameter server different tensorflow installed on worker. (e.g: different versions of tensorflow). how distributed tensorflow reconcile potential differences in execution code between workers , parameter servers?
i'd it's highly discouraged. depending on how many versions differ, in cases, errors. in base cases, things seemingly run e.g. nothing trains , you'll have hard time figuring out why....
Comments
Post a Comment