docs(inference): clarify 64 GB DGX Spark evaluation and releases

[miyoungc]

Outcome

64 GB DGX Spark users can find the Experimental Qwen profile's memory requirements and select a release containing both fixes. The docs distinguish reported physical OpenClaw validation from the hosted installer and broader qualification.

Reason

The vLLM page and v0.0.131 changelog still said physical qualification was pending without recording the bounded hardware evidence in PR #12633. The default installer selects lkg, currently v0.0.124, which lacks the fixes in PR #12502 and #12633.

Changes

• Clarify the 60,000,000,000-byte total unified-memory admission floor and separate free-memory preflight.
• Retain Experimental status and describe the reported source-build OpenClaw test and its limits.
• Identify v0.0.131 for evaluation, automatic Express option 1, and the fixed option 2 capacity limit.
• Link hardware prerequisites to the owning vLLM page and correct the v0.0.131 entry's validation statement.

Verification

• npm run docs — passed; zero errors, two warnings, 125 published pages and changelog routes checked.
• Inspected generated OpenClaw, Hermes, and Deep Agents pages; shared text names OpenClaw as the tested agent.
• npm run catalog:compile and focused vLLM selection, installation, and memory-preflight tests — 59 tests passed across three files.
• Normal pre-commit and commit-msg hooks passed; normal pre-push publication validation passed.
• git diff --check passed. The diff contains no secrets, API keys, or credentials. This documentation-only correction adds no runtime behavior or tests.

Review notes

No hardware test ran for this documentation change. PR #12633 reports one physical 64 GB DGX Spark source build, existing swap, automatic OpenClaw onboarding, local chat, read/write/exec tools, a 30,037-token prompt, and one restart. The hosted installer, prolonged use, and concurrent workloads were not tested. The profile remains Experimental; this PR does not promote its catalog validation level.

Documentation Writer Review

• Independent documentation writer reviewed the completed change.
• Result: docs-updated.
• Surface: Codex Desktop.
• Evidence: docs/inference/set-up-vllm.mdx, docs/get-started/prerequisites.mdx, and docs/changelog/2026-10-06.mdx; writing rules, documentation style, source accuracy, generated variants, routes, and validation reviewed.

---

Signed-off-by: Miyoung Choi miyoungc@nvidia.com

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

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

[github-actions]

🌿 Preview your docs: https://nvidia-preview-pr-12780.docs.buildwithfern.com/nemoclaw

[github-actions]

PR Review Advisor finished for commit 5bf0417. Include the Advisor findings in the complete PR feedback collection. Verify and group valid findings before repair.

Request review only when Require no Advisor blockers is green.

All previous runs