hi-ml icon indicating copy to clipboard operation
hi-ml copied to clipboard

Hyperdrive jobs can't be submitted with SDK v2

Open ant0nsc opened this issue 2 years ago • 2 comments

ant0nsc avatar Apr 13 '23 20:04 ant0nsc

the docker shm size is not propagated to the child runs, so they get 2Gb (default) and immediately go out of memory (dataloader killed error)

ant0nsc avatar May 03 '23 16:05 ant0nsc

the docker shm size is not propagated to the child runs, so they get 2Gb (default) and immediately go out of memory (dataloader killed error)

FYI: https://github.com/microsoft/hi-ml/pull/880#issuecomment-2108464251

fepegar avatar May 13 '24 18:05 fepegar