fix(container): update image ghcr.io/defilantech/charts/llmkube ( 0.9.24 → 0.9.25 )

[renovate]

This PR contains the following updates:

Package	Update	Change
ghcr.io/defilantech/charts/llmkube	patch	0.9.24 → 0.9.25

---

Release Notes

defilantech/LLMKube (ghcr.io/defilantech/charts/llmkube)

v0.9.25

Compare Source

Features

• foreman: diff-coverage gate for added lines no test executes (#​1728) (9666346)
• foreman: enforce reviewer execution runs via a no-test-run rail (#​1747) (152811a)
• foreman: live turn stream so a coder run can be watched as it happens (#​1717) (34b9564)
• foreman: report missing planned children in Dispatched condition (#​1743) (e9ce0ff)
• modelrouter: AgentGateway data plane beside Proxy and Gateway (#​1711) (af497b6)
• model: send a Hugging Face token on operator-driven downloads (#​1750) (#​1758) (5ef4fa0)
• multi-node inference groups (spec.multiNode) for vLLM (#​1746) (b56c080)
• observability: pin the llama.cpp metric contract (#​1186) (#​1754) (da6dda9)

Bug Fixes

• agent: sign s3:// model fetches on the metal path (#​1709) (3f30570)
• agent: skip comment-only hunks in the per-hunk mutation pass (#​1742) (82c0ffd)
• controller: bound inference container memory with a limit, and validate the quantity at admission (#​1726) (e4a3669)
• controller: make zero-findings scope-overlap demotion inert (#​1737) (8a44529)
• foreman: bound the dependency wait when TimeoutSeconds is unset (#​1736) (db98633)
• foreman: clear Dispatched condition on terminal Workloads (#​1741) (9786684)
• foreman: pass upstreamURL to the coder post-push envtest gate (#​1732) (8d47306)
• foreman: preserve the workspace when the agent loop ends unsuccessfully (#​1716) (d17a7dd)
• foreman: report gate Job DeadlineExceeded as GATE-ERROR and make the gate deadline settable (#​1748) (#​1751) (ef69b4c)
• foreman: report rejected submissions on max-turns INCOMPLETE (#​1714) (e4f88e9)
• foreman: report the packages the diff-coverage gate could not measure (#​1734) (15ba740)
• foreman: validate resolvedBy before ALREADY-RESOLVED skips verification (#​1730) (2b35d71)

Documentation

• labs: link the build page absolutely so the site prerender resolves it (#​1756) (5543aa9)
• labs: reference build for DeepSeek V4 Flash Vision on two DGX Sparks (#​1755) (e80fd7d)

---

Configuration

📅 Schedule: (UTC)

• Branch creation
  • At any time (no schedule defined)
• Automerge
  • At any time (no schedule defined)

🚦 Automerge: Disabled by config. Please merge this manually once you are satisfied.

♻ Rebasing: Whenever PR is behind base branch, or you tick the rebase/retry checkbox.

🔕 Ignore: Close this PR and you won't be reminded about this update again.

---

• If you want to rebase/retry this PR, check this box

---

This PR was generated by Mend Renovate. View the repository job log.

[1Solon]

Reviewed the controller memory-limit change. This branch includes hostMemory 34Gi for the existing 10Gi RAM cache; live kube5 capacity remains sufficient (38.949Gi projected requests / 43.598Gi allocatable). Server-side dry runs passed for the model, InferenceService, OCIRepository and HelmRelease. Rollout and local inference verification will follow before #2289.

[1Solon]

Post-merge verification passed at c77ee7e: controller 0.9.25 ready; model/InferenceService ready; actual pod request and limit both 34Gi, cgroup memory.max=36507222016. Loopback chat returned OK on b10775-67a17c17c (17 prompt / 2 completion tokens; 2/2 MTP draft accepted). Zero restarts and all cgroup OOM/max counters zero. Verified before merging #2289.