Description
Issue summary:
In docs/inference/inference-options.md, Provider Status table incorrectly states that Local vLLM requires both NEMOCLAW_EXPERIMENTAL=1 and an already-running server on localhost:8000.
https://docs.nvidia.com/nemoclaw/latest/inference/inference-options.html#provider-status
Actual:
Expected:
The Provider Status docs should distinguish the two paths: already-running vLLM appears when detected on localhost:8000; managed vLLM install/start requires NEMOCLAW_EXPERIMENTAL=1 or explicit NEMOCLAW_PROVIDER=install-vllm.In v0.0.44, an already-running vLLM server is offered in the onboarding provider list without requiring NEMOCLAW_EXPERIMENTAL=1.
The experimental flag, or NEMOCLAW_PROVIDER=install-vllm, is only required for the managed install/start path that pulls images and starts a container.
Reference of code logic:
Here’s the v0.0.44 logic.
EXPERIMENTAL is only true when NEMOCLAW_EXPERIMENTAL=1:
const EXPERIMENTAL = process.env.NEMOCLAW_EXPERIMENTAL === 1 ;
Already-running vLLM is offered regardless of EXPERIMENTAL:
const explicitProvider = (process.env.NEMOCLAW_PROVIDER || ).trim().toLowerCase();
const userChoseManagedVllm = explicitProvider === install-vllm ;
if (vllmRunning) {
options.push({
key: vllm ,
label: `Local vLLM [experimental] (localhost:${VLLM_PORT}) — running (suggested)`,
});
} else if (vllmProfile && (userChoseManagedVllm || EXPERIMENTAL)) {
const verb = hasVllmImage ? Start : Install ;
options.push({ key: install-vllm , label: `${verb} vLLM (${vllmProfile.name})` });
}Managed install/start is only offered when vLLM is not running and either NEMOCLAW_PROVIDER=install-vllm or NEMOCLAW_EXPERIMENTAL=1 is set:
} else if (vllmProfile && (userChoseManagedVllm || EXPERIMENTAL)) {
const verb = hasVllmImage ? Start : Install ;
options.push({ key: install-vllm , label: `${verb} vLLM (${vllmProfile.name})` });
}If the user selects managed install/start, it calls installVllm(...), then intentionally falls through to the existing-running vLLM path:
} else if (selected.key === install-vllm ) {
// ...
const result = await installVllm(vllmProfile, {
hasImage: hasVllmImage,
nonInteractive: isNonInteractive(),
promptFn: prompt,
});
// ...
selected = { key: vllm , label: `Local vLLM (localhost:${VLLM_PORT}) — running` };
}
if (selected.key === vllm ) {
console.log(` ✓ Using existing vLLM on localhost:${VLLM_PORT}`);
provider = vllm-local ;
// queries /v1/models, validates endpoint, then forces chat completions
}Bug Details
| Field |
Value |
| Priority |
Unprioritized |
| Action |
Dev - Open - To fix |
| Disposition |
Open issue |
| Module |
Machine Learning - NemoClaw |
| Keyword |
NemoClaw, NemoClaw_Docs, NEMOCLAW_GH_SYNC_APPROVAL |
[NVB#6186688]
Description
Issue summary:
In docs/inference/inference-options.md, Provider Status table incorrectly states that Local vLLM requires both
NEMOCLAW_EXPERIMENTAL=1and an already-running server onlocalhost:8000.https://docs.nvidia.com/nemoclaw/latest/inference/inference-options.html#provider-status
Actual:
Expected:
The Provider Status docs should distinguish the two paths: already-running vLLM appears when detected on
localhost:8000; managed vLLM install/start requiresNEMOCLAW_EXPERIMENTAL=1or explicitNEMOCLAW_PROVIDER=install-vllm.Inv0.0.44, an already-running vLLM server is offered in the onboarding provider list without requiringNEMOCLAW_EXPERIMENTAL=1.The experimental flag, or
NEMOCLAW_PROVIDER=install-vllm, is only required for the managed install/start path that pulls images and starts a container.Reference of code logic:
Here’s the
v0.0.44logic.EXPERIMENTALis onlytruewhenNEMOCLAW_EXPERIMENTAL=1:Already-running vLLM is offered regardless of
EXPERIMENTAL:Managed install/start is only offered when vLLM is not running and either
NEMOCLAW_PROVIDER=install-vllmorNEMOCLAW_EXPERIMENTAL=1is set:If the user selects managed install/start, it calls
installVllm(...), then intentionally falls through to the existing-running vLLM path:Bug Details
[NVB#6186688]