fix(vllm): correct qwen-3.8 spec.files for the philbert layout

[Tanguille]

#4551 moved source to philbert440 but left spec.files listing cyankiwi's layout. The model-downloader init container fetches exactly those names and exit 1s on a 404, so the pod crashlooped (Init:Error, 5 restarts) as soon as the Deployment was regenerated.

Latent until now: the running Deployment carried a correct MODEL_FILES env from an earlier resolution, and the old cyankiwi weights on disk let the downloader's "kept cached copy" branch mask the wrong list. Resuming the Kustomization regenerated the Deployment and removing the stale weights took away the fallback.

Verified against the pinned revision 7908d42a:

file	content-length
model.safetensors	18,698,467,264
model-mtp.safetensors	849,400,424
model.safetensors.index.json	205,894
config.json	20,883
generation_config.json	214
tokenizer.json	12,809,320
tokenizer_config.json	17,928
chat_template.jinja	8,952
preprocessor_config.json	390
video_preprocessor_config.json	385

Removed, all 404 at that revision: model-0000{1..5}-of-00005.safetensors, merges.txt, vocab.json (philbert folds merges/vocab into tokenizer.json).

Deleting and recreating the Model CR does not help — the operator reproduces the list straight from spec.files, so the fix has to be in git.

[coderabbitai]

Warning

Review limit reached

@Tanguille, you've reached your PR review limit, so we couldn't start this review.

Next review available in: 26 minutes

Limit details: You’ve used the included review currently available.

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

How can I continue?

Wait for the limit to reset, then comment @coderabbitai review or push new commits to the PR.

An organization admin can change what happens after included review limits in Billing.

How do review limits work?

CodeRabbit enforces per-developer PR review limits within each organization.

For paid Pro and Pro+ reviews, CodeRabbit uses a developer's included PR review attempts over the past 7 days to set the current hourly allowance. At typical activity levels, the full plan allowance applies. Higher sustained activity can lower the allowance until earlier attempts leave the 7-day window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 2654f7b9-c4bb-4caa-bb87-91cfa351aca7

📥 Commits

Reviewing files that changed from the base of the PR and between aa60581 and aadc79f.

📒 Files selected for processing (1)

• kubernetes/apps/ai/llmkube/models/qwen38-27b-vllm.yaml

---

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.}

[tanguille-cluster]

@@ spec.files @@
# inference.llmkube.dev/v1alpha1/Model/ai/qwen38-27b-vllm
! - seven list entries removed:
- - model-00001-of-00005.safetensors
- - model-00002-of-00005.safetensors
- - model-00003-of-00005.safetensors
- - model-00004-of-00005.safetensors
- - model-00005-of-00005.safetensors
- - merges.txt
- - vocab.json
! + two list entries added:
+ - model.safetensors
+ - model-mtp.safetensors

[github-actions]

AI Automated Review

Analysis engine: omniroute@http://litellm.ai.svc.cluster.local/v1 (openai)

Recommendation: Approve.

The change correctly updates spec.files to match the philbert440 layout, resolving the crashloop caused by the mismatch introduced in PR 4551. The file list aligns with the verified content-lengths provided in the PR description, and the explanatory comment is appropriate for this repository's conventions.

Standards Compliance

• Conventions: Follows existing patterns; changes are small and focused.
• Validation: No shell scripts touched, so shellcheck validation is not applicable.

Unknowns or Needs Verification

• Live Cluster State: While the PR body states the pod was crashlooping, I have not independently verified the current cluster state or the specific revision 7908d42a of the model source. However, the logic is sound based on the provided evidence.