Skip to content

Organize workloads into model directories#47

Draft
khluu wants to merge 1 commit into
mainfrom
codex/organize-workloads-by-model
Draft

Organize workloads into model directories#47
khluu wants to merge 1 commit into
mainfrom
codex/organize-workloads-by-model

Conversation

@khluu

@khluu khluu commented Jul 22, 2026

Copy link
Copy Markdown
Member

This PR was authored with assistance from Codex.

Summary

  • move all 18 existing workload recipes into workloads/<model>/<hardware>.yaml
  • recursively discover nested workloads and fix profile lookup for nested parser paths
  • support canonical model/hardware selectors while preserving YAML names and former flat filename aliases
  • update README/agent examples and add discovery/selection regression coverage

Local validation

  • parsed all 18 nested workloads with the documented stubbed lm_eval registry
  • python3 .buildkite/test_generate_pipeline.py (10/10 passed)
  • generated the full nightly pipeline successfully
  • generated targeted pipelines with both qwen3_5/h200 and legacy deepseek_v4_pro_5_h200 selectors
  • bash -n lib/run.sh lib/server.sh lib/run_lm_eval.sh
  • git diff --check

Coordination

PRs #43-#46 stay independently testable against the current flat layout. Before this directory PR lands, refresh it after those model PRs so their new B200 files are moved into the corresponding model directories too.

Move hardware recipes under per-model directories, teach the pipeline and parser about nested paths, preserve legacy WORKLOADS selectors, and update docs and regression coverage.

Co-Authored-By: Codex <noreply@anthropic.com>
Signed-off-by: khluu <khluu000@gmail.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