cicd

Diagnose CI failures and manage the PR workflow for NeMo AutoModel pipelines.

960|316|Updated May 21, 2025
One-click install
npx skills add https://github.com/NVIDIA-NeMo/Automodel --skill cicd-nvidia-nemo
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: cicd
Source: https://github.com/NVIDIA-NeMo/Automodel/tree/main/.agents/contributor-skills/cicd
Command: npx skills add https://github.com/NVIDIA-NeMo/Automodel --skill cicd-nvidia-nemo

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Contributors to NeMo AutoModel often struggle to understand why CI did not run, how the copy-pr-bot trust mechanism works, and how to locate and investigate failing pipeline jobs without digging through large logs blindly. ## Core Features & Use Cases - Commit and PR Conventions: Enforces DCO sign-off, Conventional Commits PR titles, and branch naming rules so validation checks pass. - CI Triggering Explained: Documents how copy-pr-bot gates CI on GPG-signed commits or /ok to test <sha> comments from NVIDIA members. - Failure Investigation Workflow: Provides gh CLI commands to map a pull-request/<number> branch back to its PR, fetch failing job logs, and scan them in chunks. - Use Case: A contributor pushes a commit but CI never starts; the Skill explains the trust mechanism and instructs posting /ok to test <full-sha> on the PR. ## Quick Start Ask the assistant to investigate why CI is red on my NeMo AutoModel pull request and fetch the failing job logs.

Frequently Asked Questions about cicd

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
Why did CI not run on my NeMo AutoModel pull request?▼

CI is triggered on push, not pull_request, and copy-pr-bot only runs it for trusted code. Either all commits must be GPG-signed by a verified NVIDIA contributor, or an NVIDIA member must post `/ok to test <commit-sha>` on the PR.

How do I investigate a failing CI job on a pull-request branch?▼

Extract the PR number from the `pull-request/<number>` branch name, then use `gh pr checks` to find the failing job and `gh run view <run_id> --log-failed` to fetch logs. Scan logs in chunks with tail or sed rather than loading them whole.

How do I fix a missing DCO sign-off on a commit?▼

Amend the commit with `git commit --amend -s` to add the Signed-off-by line. All commits require DCO sign-off, so always commit with `git commit -s` going forward.

What PR title format passes the Validate PR title check?▼

Titles must follow Conventional Commits: `type(scope)?: description` with types like feat, fix, docs, refactor, test, ci, or chore, and must be 80 characters or fewer. Bracket-prefixed styles like `[model] feat: ...` fail validation.

Why do multi-GPU tests fail silently in CI?▼

Silent multi-GPU failures usually mean `CUDA_VISIBLE_DEVICES` is not set, so set it explicitly. Port conflicts from torchrun are another cause; pass `--master_port=<unused_port>` or set `MASTER_PORT`.