feat(vllm): add NEMOCLAW_VLLM_MODEL override + gated-model check#3642
Merged
Conversation
Signed-off-by: Tinson Lai <tinsonl@nvidia.com>
|
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. |
Contributor
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Enterprise Run ID: 📒 Files selected for processing (3)
🚧 Files skipped from review as they are similar to previous changes (2)
📝 WalkthroughWalkthroughThis PR implements the ChangesvLLM Model Registry and Selection
vLLM Container and Orchestration Integration
User-Facing Documentation
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes Suggested labels
Suggested reviewers
Poem
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Warning There were issues while running some tools. Please review the errors and either fix the tool's configuration or disable the tool if it's a critical failure. 🔧 ESLint
ESLint skipped: no ESLint configuration detected in root package.json. To enable, add Comment |
Contributor
E2E Advisor RecommendationRequired E2E: Dispatch hint: Auto-dispatched E2E: Full advisor summaryE2E Recommendation AdvisorBase: Required E2E
Optional E2E
New E2E recommendations
Dispatch hint
|
…ment NEMOCLAW_VLLM_MODEL Signed-off-by: Tinson Lai <tinsonl@nvidia.com>
Contributor
Selective E2E Results —
|
| Job | Result |
|---|---|
| gpu-e2e | ⏭️ skipped |
Contributor
There was a problem hiding this comment.
Actionable comments posted: 2
🧹 Nitpick comments (6)
docs/reference/commands.md (1)
1156-1156: ⚡ Quick winUse US English spelling.
"Recognised" uses British spelling.
Prefer "Recognized" for consistency with US English conventions.Suggested revision
-| `NEMOCLAW_VLLM_MODEL` | registry slug or Hugging Face model id | Selects the model the managed-vLLM install path serves. Recognised slugs: `qwen3.6-27b`, `nemotron-3-nano-4b`, `deepseek-r1-distill-70b`. Unset uses the per-platform profile default. Gated models (e.g. `deepseek-r1-distill-70b`) require `HF_TOKEN` or `HUGGING_FACE_HUB_TOKEN`. | +| `NEMOCLAW_VLLM_MODEL` | registry slug or Hugging Face model id | Selects the model the managed-vLLM install path serves. Recognized slugs: `qwen3.6-27b`, `nemotron-3-nano-4b`, `deepseek-r1-distill-70b`. Unset uses the per-platform profile default. Gated models (e.g. `deepseek-r1-distill-70b`) require `HF_TOKEN` or `HUGGING_FACE_HUB_TOKEN`. |🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@docs/reference/commands.md` at line 1156, Update the US English spelling in the table cell describing the NEMOCLAW_VLLM_MODEL environment variable: change the word "Recognised" to "Recognized" in the sentence that lists recognised slugs (e.g., `qwen3.6-27b`, `nemotron-3-nano-4b`, `deepseek-r1-distill-70b`) so the description reads "Recognized slugs: ...". Keep the rest of the text (gating note about `HF_TOKEN`/`HUGGING_FACE_HUB_TOKEN` and default behavior) unchanged.docs/inference/use-local-inference.md (5)
312-312: ⚡ Quick winUse US English spelling.
"Recognised" uses British spelling.
Prefer "Recognized" for consistency with US English conventions.Suggested revision
-Recognised slugs: +Recognized slugs:🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@docs/inference/use-local-inference.md` at line 312, Change the British spelling "Recognised slugs:" to US English "Recognized slugs:" in the docs string (look for the literal "Recognised slugs:" in use-local-inference.md) so the documentation uses consistent US English spelling.
310-310: ⚡ Quick winRemove informal conversational tone.
The phrase "by, well, default" includes an unnecessary conversational interjection.
Use direct technical language instead.Suggested revision
-Managed vLLM serves the profile default by, well, default. +Managed vLLM serves the profile default model.As per coding guidelines: documentation voice should be direct and professional, avoiding conversational asides.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@docs/inference/use-local-inference.md` at line 310, Replace the informal sentence "Managed vLLM serves the profile default by, well, default." with a direct, professional version such as "Managed vLLM serves the profile default." — edit the sentence text in the docs to remove the conversational interjection ("by, well,") and ensure the tone is direct and technical in the surrounding paragraph.
320-320: ⚡ Quick winSplit into one sentence per line.
Multiple sentences appear on the same line.
Suggested revision
-The slug is case-insensitive; the full Hugging Face id is also accepted. +The slug is case-insensitive. +The full Hugging Face id is also accepted.As per coding guidelines: use one sentence per line in source to make diffs readable.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@docs/inference/use-local-inference.md` at line 320, The line "The slug is case-insensitive; the full Hugging Face id is also accepted." contains two sentences on one line; break them into separate sentences each on its own line (e.g., "The slug is case-insensitive." on one line and "The full Hugging Face id is also accepted." on the next) and replace the semicolon with a period so the source follows the one-sentence-per-line guideline.
311-311: ⚡ Quick winSplit into one sentence per line.
Multiple sentences appear on the same line.
This makes diffs harder to read.Suggested revision
-Export `NEMOCLAW_VLLM_MODEL=<slug>` before invoking the installer to swap in a different model from the registry; NemoClaw uses the matching `vllm serve` flags (reasoning parser, tool-call parser, `--max-model-len`). +Export `NEMOCLAW_VLLM_MODEL=<slug>` before invoking the installer to swap in a different model from the registry. +NemoClaw uses the matching `vllm serve` flags (reasoning parser, tool-call parser, `--max-model-len`).As per coding guidelines: use one sentence per line in source to make diffs readable.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@docs/inference/use-local-inference.md` at line 311, The line containing "Export `NEMOCLAW_VLLM_MODEL=<slug>` before invoking the installer to swap in a different model from the registry; NemoClaw uses the matching `vllm serve` flags (reasoning parser, tool-call parser, `--max-model-len`)." has multiple sentences on one line—split it so each sentence is its own line, preserving the inline code spans (`NEMOCLAW_VLLM_MODEL=<slug>`, `vllm serve`, and `--max-model-len`) and punctuation, e.g., one line for the export instruction, one line explaining NemoClaw uses matching vllm serve flags, and one line listing the specific flags.
334-334: ⚡ Quick winSplit into one sentence per line.
Multiple sentences appear on the same line.
Suggested revision
-The token check runs on the host before any docker pull, so a missing or empty token aborts onboarding before bandwidth is spent on a 401. +The token check runs on the host before any docker pull. +A missing or empty token aborts onboarding before bandwidth is spent on a 401.As per coding guidelines: use one sentence per line in source to make diffs readable.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@docs/inference/use-local-inference.md` at line 334, Split the long sentence in docs/inference/use-local-inference.md into two separate sentences each on its own line: change "The token check runs on the host before any docker pull, so a missing or empty token aborts onboarding before bandwidth is spent on a 401." into two sentences (e.g., "The token check runs on the host before any docker pull." and "A missing or empty token aborts onboarding before bandwidth is spent on a 401.") so each sentence occupies its own source line for clearer diffs.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@src/lib/inference/vllm-models.test.ts`:
- Around line 6-12: The test is importing compiled output; change the import in
vllm-models.test.ts to point at the source module so tests reflect source edits
— replace the import from "../../../dist/lib/inference/vllm-models" with the
corresponding source path (e.g. "../../../src/lib/inference/vllm-models") so
DEFAULT_VLLM_MODEL, VLLM_MODELS, assertGatedModelAccess, buildVllmServeCommand,
and selectVllmModelFromEnv are loaded from the source files.
In `@src/lib/inference/vllm-models.ts`:
- Line 79: The Nemotron model entry in src/lib/inference/vllm-models.ts has
maxModelLen set to 262000 (a typo); update the property maxModelLen on the
Nemotron model object to 262144 to match vLLM documentation and the Qwen
registry entry so the model config uses the correct 256K token length.
---
Nitpick comments:
In `@docs/inference/use-local-inference.md`:
- Line 312: Change the British spelling "Recognised slugs:" to US English
"Recognized slugs:" in the docs string (look for the literal "Recognised slugs:"
in use-local-inference.md) so the documentation uses consistent US English
spelling.
- Line 310: Replace the informal sentence "Managed vLLM serves the profile
default by, well, default." with a direct, professional version such as "Managed
vLLM serves the profile default." — edit the sentence text in the docs to remove
the conversational interjection ("by, well,") and ensure the tone is direct and
technical in the surrounding paragraph.
- Line 320: The line "The slug is case-insensitive; the full Hugging Face id is
also accepted." contains two sentences on one line; break them into separate
sentences each on its own line (e.g., "The slug is case-insensitive." on one
line and "The full Hugging Face id is also accepted." on the next) and replace
the semicolon with a period so the source follows the one-sentence-per-line
guideline.
- Line 311: The line containing "Export `NEMOCLAW_VLLM_MODEL=<slug>` before
invoking the installer to swap in a different model from the registry; NemoClaw
uses the matching `vllm serve` flags (reasoning parser, tool-call parser,
`--max-model-len`)." has multiple sentences on one line—split it so each
sentence is its own line, preserving the inline code spans
(`NEMOCLAW_VLLM_MODEL=<slug>`, `vllm serve`, and `--max-model-len`) and
punctuation, e.g., one line for the export instruction, one line explaining
NemoClaw uses matching vllm serve flags, and one line listing the specific
flags.
- Line 334: Split the long sentence in docs/inference/use-local-inference.md
into two separate sentences each on its own line: change "The token check runs
on the host before any docker pull, so a missing or empty token aborts
onboarding before bandwidth is spent on a 401." into two sentences (e.g., "The
token check runs on the host before any docker pull." and "A missing or empty
token aborts onboarding before bandwidth is spent on a 401.") so each sentence
occupies its own source line for clearer diffs.
In `@docs/reference/commands.md`:
- Line 1156: Update the US English spelling in the table cell describing the
NEMOCLAW_VLLM_MODEL environment variable: change the word "Recognised" to
"Recognized" in the sentence that lists recognised slugs (e.g., `qwen3.6-27b`,
`nemotron-3-nano-4b`, `deepseek-r1-distill-70b`) so the description reads
"Recognized slugs: ...". Keep the rest of the text (gating note about
`HF_TOKEN`/`HUGGING_FACE_HUB_TOKEN` and default behavior) unchanged.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Enterprise
Run ID: e9140b1a-2d1c-422d-b8f7-25a170058be6
📒 Files selected for processing (6)
docs/inference/use-local-inference.mddocs/reference/commands.mdsrc/lib/inference/vllm-models.test.tssrc/lib/inference/vllm-models.tssrc/lib/inference/vllm.tstest/detect-vllm-profile.test.ts
…el_len to 262144 Signed-off-by: Tinson Lai <tinsonl@nvidia.com>
Contributor
Selective E2E Results —
|
| Job | Result |
|---|---|
| gpu-e2e | ⏭️ skipped |
cv
approved these changes
May 16, 2026
Contributor
Selective E2E Results —
|
| Job | Result |
|---|---|
| gpu-e2e | ⏭️ skipped |
12 tasks
ericksoa
pushed a commit
that referenced
this pull request
May 18, 2026
## Summary Updates the NemoClaw documentation for the v0.0.45 release by summarizing the user-facing changes merged since v0.0.44 and bumping the docs version metadata. Refreshes generated user skills so agent-facing references match the source docs. ## Changes - Added v0.0.45 release notes covering onboarding recovery, local inference, channel cleanup, share mount diagnostics, uninstall cleanup, and security redaction updates. - Updated command and troubleshooting docs for sandbox name limits, GPU gateway reuse, DNS preflight behavior, channel removal cleanup, and share mount path validation. - Bumped docs version metadata to 0.0.45 and regenerated NemoClaw user skills from the docs. - Source summary: #3672 -> `docs/reference/commands.md`: documented channel removal detaching bridge providers and un-applying channel policy presets. - Source summary: #3678 -> `docs/about/release-notes.md`: documented Ollama streamed usage accounting in the release notes. - Source summary: #3670 -> `docs/reference/commands.md`, `docs/reference/troubleshooting.md`: documented safe GPU gateway replacement behavior. - Source summary: #3664 -> `docs/about/release-notes.md`: documented blueprint permission normalization in the release notes. - Source summary: #3181 -> `docs/reference/troubleshooting.md`: documented GPU toolkit guidance when host drivers work but passthrough is disabled. - Source summary: #3554 -> `docs/about/release-notes.md`: documented host `openshell-gateway` cleanup during uninstall. - Source summary: #3651 -> `docs/reference/troubleshooting.md`: documented the uncached `.invalid` DNS preflight probe. - Source summary: #3643 -> `docs/reference/commands.md`: included existing `NEMOCLAW_PROVIDER` interactive-mode behavior in generated docs. - Source summary: #3647 -> `docs/reference/commands.md`: documented remote sandbox path verification for `share mount`. - Source summary: #3646 -> `docs/reference/commands.md`: included existing local writable mount target guidance in generated docs. - Source summary: #3642 -> `docs/inference/use-local-inference.md`, `docs/reference/commands.md`: documented managed-vLLM model override and gated-model token checks. - Source summary: #3639 -> `docs/reference/commands.md`: documented the 63-character sandbox name limit. ## Type of Change - [ ] Code change (feature, bug fix, or refactor) - [ ] Code change with doc updates - [ ] Doc only (prose changes, no code sample modifications) - [x] Doc only (includes code sample changes) ## Verification - [ ] `npx prek run --all-files` passes - [ ] `npm test` passes - [ ] Tests added or updated for new or changed behavior - [x] No secrets, API keys, or credentials committed - [x] Docs updated for user-facing behavior changes - [x] `make docs` builds without warnings (doc changes only) - [x] Doc pages follow the [style guide](https://github.com/NVIDIA/NemoClaw/blob/main/docs/CONTRIBUTING.md) (doc changes only) - [ ] New doc pages include SPDX header and frontmatter (new pages only) Commit hooks passed for the staged files. A standalone `npx prek run --all-files` attempt was blocked by sandbox access to `/Users/miyoungc/.cache/prek/prek.log`, so that checkbox is left unchecked. --- <!-- DCO sign-off required by CI. Run: git config user.name && git config user.email --> Signed-off-by: Miyoung Choi <miyoungc@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **Documentation** * Enhanced CLI command reference documentation with clearer guidance on onboarding, GPU passthrough, inference configuration, channel removal, and shared mounts. * Improved troubleshooting sections with better DNS resolution and GPU passthrough remediation steps. * Added documentation for overriding managed vLLM model selection. * Updated release notes for v0.0.45 reflecting infrastructure and workflow improvements. * **Version Bump** * Released v0.0.45. <!-- review_stack_entry_start --> [](https://app.coderabbit.ai/change-stack/NVIDIA/NemoClaw/pull/3755?utm_source=github_walkthrough&utm_medium=github&utm_campaign=change_stack) <!-- review_stack_entry_end --> <!-- end of auto-generated comment: release notes by coderabbit.ai -->
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds a scoped escape hatch on the express vLLM install path: setting
NEMOCLAW_VLLM_MODEL=<slug>picks a different entry from a small new registry, swaps in the rightvllm serveflags, and fails fast when a gated model is requested without a Hugging Face token (which is now forwarded into the managed-vLLM container sohf downloadandvllm servecan authenticate). An interactive model picker on the express path is not part of this PR — that remains a separate design item; the env-var path covers the gated-model coverage QA was trying to exercise.Related Issue
Fixes #3566
Fixes #3572
Changes
src/lib/inference/vllm-models.tswith a typed registry, env-var resolver, gated-access check, and a sharedbuildVllmServeCommandthat merges common flags with the model-specific args.nullwhenNEMOCLAW_VLLM_MODELis unset so the caller can keep the per-platform profile default (the generic-Linux profile must stay on Nemotron-Nano-4B for VRAM headroom).VllmProfileto carry adefaultModel: VllmModelDefinstead of staticmodel/commandstrings;installVllmresolves the model per call, validates gated access, builds the serve command on the fly, and forwardsHF_TOKEN/HUGGING_FACE_HUB_TOKENinto both thehf downloadone-shot and the long-livedvllm servecontainer using the-e KEY(key-only) form so the secret stays out of the docker argv; the value is added back via the runner'senvoption so docker can inherit it without breaking the subprocess-env allowlist.use-local-inference.md; add it to the env-var reference table incommands.mdto satisfy the env-var doc gate.detect-vllm-profile.test.tsto assert againstprofile.defaultModel.idand add coverage forbuildHfTokenDockerArgs+buildHfTokenForwardEnv(HF_TOKEN preferred, HUGGING_FACE_HUB_TOKEN fallback, empty/whitespace handled).vllm-models.test.tscovering the env-var resolver (slug + HF id forms, unset-returns-null), unknown-value error, gated-access check, and command-builder output.Type of Change
Verification
Signed-off-by: Tinson Lai tinsonl@nvidia.com
Summary by CodeRabbit
Release Notes
New Features
Documentation