You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I was getting this error:
```
facebook.hpc_scheduler.hpcscheduler.types.HpcSchedulerServiceException:
WarmStorageSpec in task group 'torchrun' has a non-empty directory field
('checkpoint/infra') but has missing clusters field. If job requires
warm storage cluster mount, please configure storage mount options by
specifying region or conda storage mount environment variables. For more
details please see wiki:
https://www.internalfb.com/code/fbsource/fbcode/conda/mast/mount/README.md
.If you need further assistance, please make a post in
https://fb.workplace.com/groups/mast.users., error:
HpcSchedulerErrorCode::INVALID_JOB_DEFINITION (errorCode: 402)
```
Looking at the wiki, I guessed that probably I was missing the 'cluster'
info in the FUSE_SRC var, and I also noted that we already specify this
var in mount.py in torchtitan, so I copied the value from there to
mast.py and it works.
0 commit comments