Repository navigation
Conversation
Signed-off-by: Serge Panev <spanev@nvidia.com>
Contributor
Author
|
Not upstreaming the cluster harness. The Slurm bootstrap, model-server submission and continuation scripts stay alongside the rest of the private orchestration, and the scheduling cookbook points at them there instead. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stacked on #2162, which was reduced to the runner, validator and dataset prep that recent runs actually exercise.
This adds the rest of the cluster harness:
slurm/bootstrap, model server submission, the continuation gate/monitor, resume-state validation, and the GLM model profiles.None of it is exercised by the collocated rollout path used today, so it is separated for its own review. Two known issues to settle here rather than in #2162:
submit_model_server.shrunsenv -i, which strips everyGDPVAL_*a model profile exports, then hardcodes--num-nodes 2 --tp-size 8— this makesmodels/*.envinert.slurm/is a non-collocated orchestration path that has not been run recently.Merge after #2162.