The scheduler killed the job for exceeding the memory it requested
On a cluster you get exactly the memory you asked for, not what the node has. Going over gets the job killed immediately — the tool's own output stops mid-sentence and explains nothing, because it was never told.
Ask for more for that specific step, then measure rather than guess: seff <jobid> (Slurm) reports what the job actually used, which is how you set the next request. In a workflow, raise the memory directive for that one process rather than the whole run.