chore: Use self-hosted runners via repo action variable toggle

[ndizazzo]

Summary by CodeRabbit

• Changes
  • CUDA-related CI and release jobs can now switch between self-hosted GPU runners and GitHub-hosted runners based on a repository setting.
  • Linux CUDA and ARM64 build/test jobs now use conditional runner selection, improving flexibility across environments.
  • Updated job labels and comments to better reflect the current runner behavior.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 6d4e7d73-b411-4e53-9a72-d9aa2f5cd6a7

📥 Commits

Reviewing files that changed from the base of the PR and between 232c85a and 011fec6.

📒 Files selected for processing (3)

• .github/workflows/ci.yml
• .github/workflows/pr_builds.yml
• .github/workflows/release.yml

---

📝 Walkthrough

Walkthrough

Three GitHub Actions workflow files (ci.yml, pr_builds.yml, release.yml) are updated to replace hardcoded self-hosted NVIDIA/ARM64 runner labels with conditional runs-on expressions controlled by the vars.USE_SELF_HOSTED repository variable. Header comments documenting this behavior are added to each workflow.

Changes

Conditional USE_SELF_HOSTED Runner Routing

Layer / File(s)	Summary
ci.yml: CUDA job conditional routing .github/workflows/ci.yml	Adds header comment; changes linux_cuda runs-on from a fixed self-hosted label array to a conditional expression using vars.USE_SELF_HOSTED, falling back to ubuntu-24.04.
pr_builds.yml: matrix CUDA routing .github/workflows/pr_builds.yml	Adds header comment; updates linux_targets runs-on to select matrix.self_hosted_runs_on when USE_SELF_HOSTED is true; adds self_hosted_runs_on and updates runs_on in the CUDA slim matrix include.
release.yml: CUDA and ARM64 conditional routing .github/workflows/release.yml	Adds header comment; converts runs-on for build_native_runtime_linux_x86_64_cuda, smoke_linux_arm64_artifact (also renamed), and build_linux_cuda to conditional fromJson(...) expressions based on vars.USE_SELF_HOSTED.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• Mesh-LLM/mesh-llm#923: Adjusts runs-on and runner-label selection for the same CUDA-related workflow jobs (linux_cuda, linux_targets, release CUDA paths).
• Mesh-LLM/mesh-llm#924: Switches the same CUDA job self-hosted runner labels from carrack to amd64/gpu-nvidia, directly overlapping the label values used in this PR's conditional expressions.

Suggested reviewers

• michaelneale

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title matches the main change: toggling workflow runner selection between self-hosted and GitHub-hosted via a repo variable.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch chore/use-self-hosted-runner-toggle

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}