Skip to content

fix(env): address managed lockfile review follow-ups - #5604

Merged
ko3n1g merged 1 commit into
mainfrom
ko3n1g/fix/lockfile-review-followup
Nov 4, 2026
Merged

ko3n1g merged 1 commit into
mainfrom
ko3n1g/fix/lockfile-review-followup

Conversation

@ko3n1g

@ko3n1g ko3n1g commented Nov 4, 2026 •

Copy link
Copy Markdown
Contributor

Background

Follow up on the four remaining requests in Kajal's review of #4837. The merged packaging hook replaced directly consumed runtime locks with placeholders, the translation locks forced COMET outside its declared dependency limits, lock freshness was outside the required CI gate, and contributor instructions still advertised an unsupported uv version. This review supplies the issue context; a separate issue is unnecessary.

What changed

  • Preserve the MolOpt and OSWorld optional-runtime locks in the wheel so their existing uv sync --locked installers can read them. Other component locks continue to use the compressed archive. Add packaging regression coverage.
  • Remove the Transformers and Hub overrides from WMT and LongMT. Record a Transformers <5 exception in the component policy and regenerate those two locks. Test effective component constraints and COMET's supported version ranges.
  • Make the Linux/macOS lock workflow reusable from trusted CICD. Nemo_CICD_Test depends on it and rejects failure, cancellation, missing results, and unexpected skips, including docs-only runs. Include it in failure summaries. Remove its duplicate standalone main push trigger.
  • Update CONTRIBUTING, the full-suite workflow comment, and the development guide to require uv 0.13.4+, with upgrade instructions.

Details

The old Transformers 5.22.1 override reproduces a COMET inference failure: ValueError: not enough values to unpack (expected 3, got 2) in XLMREncoder.forward. Both new CPU inference tests reproduce this failure under the previous override and pass with each corrected lock.

Component Transformers Hugging Face Hub
wmt_translation 4.57.6 0.36.2
longmt_eval 4.51.3 0.36.2

The shared Transformers 5 security floor is explicitly waived for these two isolated COMET 2.2.7 environments to restore its declared API contract. This exception is recorded in scripts/component-lock-policy.json and the development guide. Existing uv installations must be upgraded before syncing; upgrading uv itself requires no lock regeneration.

flowchart LR
  C[Trusted CICD] --> L[Component lock workflow]
  L --> U[Linux: all managed locks]
  L --> M[macOS: all managed locks]
  U --> G[Nemo_CICD_Test]
  M --> G
  T[Existing required test jobs] --> G
Loading

Tested

  • Signed commit: af0196eca96c51984480f8e42b399e47eb86b727, based on merged main at 1203b0be210f8fd19ab3fc4cdc5262feb73ffed9. Signature and DCO sign-off verified.

  • pre-commit run --all-files: passed. git diff --cached --check: passed.

  • python scripts/component_locks.py --jobs 4: 267 projects, 0 failed.

  • Root regression suite: 127 passed, 1 skipped. The skip is the optional COMET module in the root environment; its two tests pass separately in each component environment.

pytest -q -o addopts='' --tb=short \
  tests/unit_tests/test_wheel_locks.py \
  tests/unit_tests/test_ci_environment.py \
  tests/unit_tests/test_component_dependencies.py \
  tests/unit_tests/test_locked_env.py \
  tests/unit_tests/test_comet_compatibility.py
  • Translation tests, from each component directory with its own venv activated: HF_HUB_OFFLINE=1 TRANSFORMERS_OFFLINE=1 OMP_NUM_THREADS=1 python -m pytest -q -o addopts='' --tb=short tests: WMT 118 passed; LongMT 45 passed.
  • Both component venvs were installed from their frozen exported locks, using matching CPU-only torch==2.12.0+cpu and torchvision==0.27.0+cpu for local validation. Run the following from the repo root with each component's Python:
HF_HUB_OFFLINE=1 TRANSFORMERS_OFFLINE=1 OMP_NUM_THREADS=1 \
  resources_servers/wmt_translation/.venv/bin/python -m pytest \
  -q -o addopts='' --tb=short tests/unit_tests/test_comet_compatibility.py
HF_HUB_OFFLINE=1 TRANSFORMERS_OFFLINE=1 OMP_NUM_THREADS=1 \
  resources_servers/longmt_eval/.venv/bin/python -m pytest \
  -q -o addopts='' --tb=short tests/unit_tests/test_comet_compatibility.py

2 passed in each environment. These construct small local XLM-R encoders and run actual referenceless and XCOMET prediction on CPU without downloading pretrained weights. Both tests fail under the previous Transformers 5.22.1 override.

  • Verified the packaging, version-range, and CI-gate regressions against the merged implementation: 9 expected failures, then all pass with this change.
  • uv build --wheel --out-dir /tmp/gym-4837-followup-dist: passed. Extracted the two runtime projects from that wheel and ran uv sync --locked --dry-run --inexact --no-dev --project <extracted-project> --python <version>: both passed, using Python 3.11 for MolOpt and 3.13.14 for OSWorld. This validates lock acceptance without installing optional runtime packages.
  • actionlint -shellcheck='' -ignore 'label ".+" is unknown' .github/workflows/component-locks.yml .github/workflows/cicd-main.yml .github/workflows/full-test-suite.yml: passed. Shell diagnostics and existing custom runner-label warnings are excluded from this syntax/dependency validation.
  • From fern/, npm run check -- --warnings: 0 errors; the authenticated redirects check is skipped because no Fern token is configured locally.
  • Model rollouts: N/A for packaging, dependency metadata, CI, and contributor documentation. The CPU compatibility tests exercise inference APIs; benchmark score validation was not run.

CI triage

  • Latest-head CICD run 41859920268, commit af0196eca96c51984480f8e42b399e47eb86b727: Both Linux/macOS component-lock checks and sandbox tests passed. All four core and all eight server test steps completed successfully. GitHub still reports several finished jobs as in progress, while combined coverage, the unit/server aggregates, and Nemo_CICD_Test remain queued without assigned runners. Those aggregate verdicts remain pending.
  • The automated source review completed with no findings. It is a comment-only review, not a formal approval.
  • Excluded as unrelated, per the requested scope:
    • Kilocode conformance fails the model_error scenario. The same failure occurs on current main at 1203b0be210f8fd19ab3fc4cdc5262feb73ffed9 in job 126670755114; its publish/check aggregates reflect that failure. The main baseline's required CI gate and all unit/server jobs passed.
    • Fern preview passes configuration validation, then Fern 5.50.4 fails parsing the generated orchestration/executors/judge_profile.mdx library page. libraries.nemo-gym.input.git reads the library from the upstream default branch. The preview workflow, library-input configuration, and judge_profile.py are unchanged from main (git diff origin/main HEAD -- .github/workflows/fern-docs-preview.yml fern/docs.yml nemo_gym/orchestration/executors/judge_profile.py is empty). This is the existing pinned preview generator's failure on main-derived source; main publication uses a different CLI path and succeeds.
  • No PR-related failure or outstanding review finding has been identified. No additional source changes were needed after CI triage. The working tree is clean.

Signed-off-by: oliver könig <okoenig@nvidia.com>
@ko3n1g
ko3n1g requested a review from a team as a code owner November 4, 2026 15:38
@ko3n1g ko3n1g added bug Something isn't working area:build Packaging, dependencies, containers, and releases needs-review PR is ready for code review and waiting on a reviewer labels Nov 4, 2026
@nemo-automation-bot
nemo-automation-bot Bot requested review from a team November 4, 2026 15:39

@svcnvidia-nemo-ci svcnvidia-nemo-ci left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

SHIP — LGTM — no concerns found.

I inspected all 15 changed files and traced the affected runtime installers, wheel lock restoration, dependency-policy merge, COMET callers, and CI result gates. The packaging hook preserves the two directly consumed runtime locks while retaining the archive path for other components. The required CI gate rejects non-successful lock results before the docs-only and nightly branches.

I couldn't retrieve one delegated report, so I completed its packaging and setup-documentation inspection directly. This was a read-only source review; I didn't execute tests or independently reproduce the reported build and CPU inference results.

Note: formal approval is not available for this request, so this review is posted as a comment only.

Review generated by codex. This was a full review. Request a full review with /review reset=true.

@ko3n1g
ko3n1g enabled auto-merge November 4, 2026 16:36
@kajalj22
kajalj22 self-requested a review November 4, 2026 16:52
@ko3n1g
ko3n1g added this pull request to the merge queue Nov 4, 2026
@github-actions

github-actions Bot commented Nov 4, 2026

Copy link
Copy Markdown
Contributor

🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA-NeMo/Gym/actions/runs/41872099263

Merged via the queue into main with commit 1a4a2c3 Nov 4, 2026
69 of 73 checks passed
@ko3n1g
ko3n1g deleted the ko3n1g/fix/lockfile-review-followup branch November 4, 2026 17:16

This branch was previously deployed

1 inactive deployment
public — af0196ec Deployed Nov 4, 2026 by copy-pr-bot[bot] via release / finalize / notify #10029
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

area:build Packaging, dependencies, containers, and releases bug Something isn't working needs-review PR is ready for code review and waiting on a reviewer

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants