I’m using wandb Launch with docker to run sweep jobs to a remote agent. The job on the remote agent seems to consistently crash at ~30 mins and appears to be working and logging correctly before that. I want to diagnose the logs from the container, but wandb Launch automatically attaches the --rm
tag to the docker run
command (deleting the container on exit). How can the --rm
tag be removed such that the container isn’t deleted upon crash so I can read the log files?
Thanks!