Skip to content

Check for checkpoints#84

Open
michaelmckinsey1 wants to merge 1 commit into
LBANN:mainfrom
michaelmckinsey1:better-checkpointing
Open

Check for checkpoints#84
michaelmckinsey1 wants to merge 1 commit into
LBANN:mainfrom
michaelmckinsey1:better-checkpointing

Conversation

@michaelmckinsey1

Copy link
Copy Markdown
Collaborator

I have been using checkpoints via

--restart \
  --run-dir /usr/workspace/mckinsey/ScaFFold-RFP/scaling/tuolumne/torchrun_hpc-scaffold_2026-06-23_06h34m11s_380741e8/benchmark_runs/benchmark_20260623-063806

and I noticed if an incorrect path is specified, e.g. if scaffold can't find a checkpoint, it will start from scratch. This led to me accidentally restarting the same jobs. I think we should check for the checkpoint file first and exit if the user made a mistake, instead of starting from the beginning and wasting compute cycles.

@michaelmckinsey1 michaelmckinsey1 self-assigned this Jul 9, 2026
@michaelmckinsey1 michaelmckinsey1 marked this pull request as ready for review July 9, 2026 19:58
@michaelmckinsey1 michaelmckinsey1 changed the title [WIP] Check checkpoints Check checkpoints Jul 9, 2026
@michaelmckinsey1 michaelmckinsey1 changed the title Check checkpoints Check for checkpoints Jul 9, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant