Skip to content

add slurm skill - #34

Open
samhita-alla wants to merge 2 commits into
flyteorg:mainfrom
samhita-alla:slurm-skill
Open

add slurm skill#34
samhita-alla wants to merge 2 commits into
flyteorg:mainfrom
samhita-alla:slurm-skill

Conversation

@samhita-alla

Copy link
Copy Markdown
Contributor

No description provided.

Signed-off-by: Samhita Alla <aallasamhita@gmail.com>
Signed-off-by: Samhita Alla <aallasamhita@gmail.com>
samhita-alla added a commit to unionai/unionai-docs that referenced this pull request Aug 7, 2026
Adds a "Migration (Slurm → Flyte 2)" group for the flyte-migrate-slurm skill
from flyteorg/flyte-agent-plugins#34, mentions Slurm in the page intro and the
skills lead-in, and bumps the skill count in the compatibility table to 21.

Framed as a guided starting point rather than a mechanical translator, and
carries over the skill's own stated limits: tightly coupled MPI simulation,
gang scheduling, and topology-aware placement stay on Slurm.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Signed-off-by: Samhita Alla <aallasamhita@gmail.com>
cosmicBboy pushed a commit to unionai/unionai-docs that referenced this pull request Aug 7, 2026
Adds a "Migration (Slurm → Flyte 2)" group for the flyte-migrate-slurm skill
from flyteorg/flyte-agent-plugins#34, mentions Slurm in the page intro and the
skills lead-in, and bumps the skill count in the compatibility table to 21.

Framed as a guided starting point rather than a mechanical translator, and
carries over the skill's own stated limits: tightly coupled MPI simulation,
gang scheduling, and topology-aware placement stay on Slurm.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Signed-off-by: Samhita Alla <aallasamhita@gmail.com>
cosmicBboy added a commit to unionai/unionai-docs that referenced this pull request Aug 7, 2026
* document the flyte-migrate-slurm skill

Adds a "Migration (Slurm → Flyte 2)" group for the flyte-migrate-slurm skill
from flyteorg/flyte-agent-plugins#34, mentions Slurm in the page intro and the
skills lead-in, and bumps the skill count in the compatibility table to 21.

Framed as a guided starting point rather than a mechanical translator, and
carries over the skill's own stated limits: tightly coupled MPI simulation,
gang scheduling, and topology-aware placement stay on Slurm.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Signed-off-by: Samhita Alla <aallasamhita@gmail.com>

* reframe the Slurm note around what ports over cleanly

The previous wording listed gang scheduling and "topology-aware placement at
the largest scales" as things to leave on Slurm, which reads as a warning to
anyone running multi-node GPU jobs. That is not the case: ClusteredTaskEnvironment
launches replicas as one JobSet, runs torchrun rendezvous, and sets the standard
RANK/WORLD_SIZE/MASTER_ADDR, so DDP and FSDP code carries over unchanged.

States that up front and narrows the caveat to what the skill actually calls
out: tightly coupled MPI simulation, and explicit InfiniBand topology control.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Signed-off-by: Samhita Alla <aallasamhita@gmail.com>

* trim the Slurm note

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Signed-off-by: Samhita Alla <aallasamhita@gmail.com>

* say Flyte 2 consistently in the intro

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Signed-off-by: Samhita Alla <aallasamhita@gmail.com>

---------

Signed-off-by: Samhita Alla <aallasamhita@gmail.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@samhita-alla

Copy link
Copy Markdown
Contributor Author

@cosmicBboy can you +1 this? it's the only remaining PR.

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