In:

https://access.redhat.com/documentation/en-us/red_hat_openshift_ai_self-managed/2.5/html/release_notes/resolved-issues_relnotes

RHOAIENG-807 - Accelerator profile toleration removed when restarting a workbench

Previously, if you created a workbench that used an accelerator profile that in turn included a toleration, restarting the workbench removed the toleration information, which meant that the restart could not complete. A freshly created GPU-enabled workbench might start the first time, but never successfully restarted afterwards because the generated pod remained forever pending. This issue is now resolved.

------> "restarted" should be replaced by "restart"