Skip to content

Add workload lint + CONTRIBUTING for pre-merge validation#41

Draft
stecasta wants to merge 1 commit into
vllm-project:mainfrom
stecasta:stecasta/perf-eval-workload-lint
Draft

Add workload lint + CONTRIBUTING for pre-merge validation#41
stecasta wants to merge 1 commit into
vllm-project:mainfrom
stecasta:stecasta/perf-eval-workload-lint

Conversation

@stecasta

Copy link
Copy Markdown

Gives perf-eval a pre-merge validation story so a broken recipe fails in review instead of at 3am on a GPU.

  • scripts/lint_workload.py: validates a recipe with no GPU (schema via parse_workload + confirms generate_pipeline emits a step). Modes: explicit paths, --changed (vs origin/main), --all.
  • Buildkite lint workloads step: automatic gate on every PR.
  • pre-commit hook + CONTRIBUTING.md documenting: lint locally, PR lint gate, then a scoped WORKLOADS= build to validate on hardware.
  • parse_workload gains SKIP_TASK_REGISTRY so the lint runs without lm_eval installed.

Validated locally: 17/17 existing recipes pass; a deliberately broken recipe (bad GPU, bogus task field) fails with a clear message.

Draft for now to gather feedback on the approach before wiring the CI gate.

AI-assisted (Claude Code).

Adds scripts/lint_workload.py (schema via parse_workload + pipeline-generation
check, no GPU required), wires it as a buildkite PR gate and a pre-commit hook,
and documents the validate-before-merge flow (lint locally, PR lint gate, then
a scoped WORKLOADS build on hardware). parse_workload gains SKIP_TASK_REGISTRY
so the lint runs without lm_eval installed.

Signed-off-by: Stefano Castagnetta <scastagnetta@nvidia.com>
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