fix(inference): harden Lightning Spark profile

docs/inference/set-up-vllm.mdx

@@ -138,6 +138,8 @@ Managed profiles and model-specific recipes use immutable image digests:
 - DGX Spark and DGX Station models without a model-specific runtime use the `linux/arm64` digest `sha256:9204569b17ee4c0eff75194b8e6e458479c8aee18953b5ab9cf359fcdac659e2` with a compressed layer size of `9.60 GB` under `nvcr.io/nvidia/vllm:26.05.post1-py3`.
 - The DGX Spark Muse Glimmer recipe uses the `linux/arm64` digest `sha256:ab0f5fc3bb81b9257a9aee801abcb0eeb94bb0523b57b2bb79349dc61e7c1e25` with a compressed layer size of `10.51 GB` under `vllm/vllm-openai`.
   It pins Hugging Face revision `d35cb79050f419c457611b1cee5c5d15b176f285` for the approximately `25.45 GB` model download.
+- The DGX Spark Nemotron 3.5 Lightning recipe uses the `linux/arm64` digest `sha256:3af90144a0926e5c5fe46ee16e5201e763dd854538b9d7ce433755f11dadaf78` with a compressed layer size of approximately `12.69 GB` under `vllm/vllm-openai`.
+  It pins Hugging Face revision `0dcd680e5585c791728c83342b311d0a0026dbeb` for the approximately `21.56 GB` model download.
 - The DGX Station Nemotron 3 Ultra express recipe uses the multi-platform index digest `sha256:0fec7ec5f3e6bc168e54899935fb0557da908a4832a1dbc88e2debcf2f889416` under `vllm/vllm-openai:v0.22.0`; on DGX Station, that index selects a `linux/arm64` manifest with a compressed layer size of `10.67 GB`.
   It also pins Hugging Face revision `183968f87ae4cedce3039313cac1fd43d112c578` for the approximately `352.38 GB` model download.
 - Generic Linux `arm64` hosts use `sha256:447995cbb57e6c7cf792cab95e9852e5f62b5fb6d2f39e030fa4eda9a54eadb4` with a compressed layer size of `9.28 GB` under `nvcr.io/nvidia/vllm:26.03.post1-py3`.
@@ -275,9 +277,11 @@ $$nemoclaw onboard --profile vllm.dgx-spark-gb10.single.nemotron-3.5-lightning-3
 ```
 
 The profile serves the public [`nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4`](https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4) checkpoint through a pinned ARM64 vLLM image.
-Its profile limits for one concurrent sequence are a 65,536-token context window, a 4,096-token batch limit, 0.7 GPU-memory utilization, FP8 KV cache, and one-token MTP speculative decoding.
+Its profile limits for one concurrent sequence are a 65,536-token context window, a 4,096-token batch limit, 0.65 GPU-memory utilization, FP8 KV cache, and one-token MTP speculative decoding.
 It uses vLLM's Python frontend because the pinned runtime's Rust frontend does not register the required `step3p5` reasoning and tool-call parsers.
-Physical single-DGX Spark validation covered plain chat, automatic and named structured tool calls, the 10-turn 32K-input/2K-output release workload, OpenClaw chat and file tools, cached restart, cleanup, and restoration of the original Qwen route.
+Physical single-DGX Spark validation of public revision `0dcd680e5585c791728c83342b311d0a0026dbeb` covered direct chat, automatic and named structured tool calls, a ten-request release workload with input lengths from 32,768 through 41,984 tokens and 2,048 output tokens per request, OpenClaw chat and file tools, and a cached restart.
+Requests 2 through 10 averaged about 102 output tokens/s.
+At 0.65 utilization, the workload completed with about 33 GiB of host memory available and no NVIDIA kernel errors.
 
 <Warning title="Experimental Nemotron 3.5 Lightning vLLM Profile">
 This profile is an explicit opt-in for one DGX Spark and remains Experimental.

managed-inference/recipes/vllm.nemotron-3.5-lightning-30b-a3b-nvfp4.spark-single.v1.yaml

@@ -56,7 +56,7 @@ spec:
       - name: --max-num-seqs
         value: 1
       - name: --gpu-memory-utilization
-        value: 0.7
+        value: 0.65
       - name: --kv-cache-dtype
         value: fp8
       - name: --moe-backend

src/lib/inference/serving/adapter-registry.test.ts

@@ -28,6 +28,8 @@ import {
 } from "./managed-cluster-topology.js";
 import type { ManagedInferenceServingRecipe } from "./types.js";
 
+const LIGHTNING_RECIPE_ID = "vllm.nemotron-3.5-lightning-30b-a3b-nvfp4.spark-single.v1";
+
 function shippedRecipe(): ManagedInferenceServingRecipe {
   const recipe = loadManagedInferenceCatalog().recipes.find(
     ({ spec }) => spec.execution.materializerRef === MANAGED_CLUSTER_VLLM_MATERIALIZER_REF,
@@ -36,9 +38,9 @@ function shippedRecipe(): ManagedInferenceServingRecipe {
   return structuredClone(recipe as ManagedInferenceServingRecipe);
 }
 
-function shippedHostLocalRecipe(): ManagedInferenceServingRecipe {
+function shippedLightningRecipe(): ManagedInferenceServingRecipe {
   const recipe = loadManagedInferenceCatalog().recipes.find(
-    ({ spec }) => spec.execution.materializerRef === HOST_LOCAL_VLLM_MATERIALIZER_REF,
+    ({ metadata }) => metadata.id === LIGHTNING_RECIPE_ID,
   );
   expect(recipe).toBeDefined();
   return structuredClone(recipe as ManagedInferenceServingRecipe);
@@ -152,23 +154,7 @@ describe("managed inference adapter registries", () => {
   });
 
   it("accepts shell-quoted JSON values for host-local vLLM arguments", () => {
-    const recipe = shippedHostLocalRecipe();
-    const structuredArgument = {
-      ...recipe,
-      spec: {
-        ...recipe.spec,
-        serve: {
-          ...recipe.spec.serve,
-          arguments: [
-            ...recipe.spec.serve.arguments,
-            {
-              name: "--speculative-config",
-              value: '{"method":"mtp","num_speculative_tokens":1}',
-            },
-          ],
-        },
-      },
-    } as ManagedInferenceServingRecipe;
+    const structuredArgument = shippedLightningRecipe();
     const unsafeArgument = {
       ...structuredArgument,
       spec: {
@@ -190,6 +176,27 @@ describe("managed inference adapter registries", () => {
     );
   });
 
+  it("does not broaden host-local vLLM environment values for structured arguments", () => {
+    const recipe = shippedLightningRecipe();
+    const structuredEnvironment = {
+      ...recipe,
+      spec: {
+        ...recipe.spec,
+        runtime: {
+          ...recipe.spec.runtime,
+          environment: {
+            ...recipe.spec.runtime.environment,
+            VLLM_TEST_CONFIG: '{"enabled":true}',
+          },
+        },
+      },
+    } as ManagedInferenceServingRecipe;
+
+    expect(getManagedInferenceRecipeRegistrationError(structuredEnvironment)).toMatch(
+      /bounded safe text/u,
+    );
+  });
+
   it("rejects schema-valid values that the registered adapter cannot execute", () => {
     const recipe = shippedRecipe();
     const unsupportedCache = {

src/lib/inference/serving/adapter-registry.ts

@@ -83,7 +83,8 @@ const PINNED_IMAGE =
 const SAFE_ENVIRONMENT_NAME = /^[A-Z][A-Z0-9_]{0,127}$/u;
 // Structured vLLM flags are passed as individual argv values after shell
 // quoting. Keep the allowlist narrow while admitting JSON objects and arrays.
-const HOST_LOCAL_SAFE_VALUE = /^[A-Za-z0-9_@%+=:,./{}[\]"-]+$/u;
+const HOST_LOCAL_SAFE_ARGUMENT_VALUE = /^[A-Za-z0-9_@%+=:,./{}[\]"-]+$/u;
+const HOST_LOCAL_SAFE_ENVIRONMENT_VALUE = /^[A-Za-z0-9_@%+=:,./-]+$/u;
 const SHA256_DIGEST = /^sha256:[0-9a-f]{64}$/u;
 const MANAGED_CLUSTER_MATERIALIZER_OWNED_ENVIRONMENT = new Set([
   "GLOO_SOCKET_IFNAME",
@@ -413,13 +414,13 @@ function validateHostLocalVllmMaterializerRecipe(
         typeof value === "string" &&
         (Buffer.byteLength(value, "utf8") > 16_384 ||
           value.includes("\0") ||
-          !HOST_LOCAL_SAFE_VALUE.test(value)),
+          !HOST_LOCAL_SAFE_ARGUMENT_VALUE.test(value)),
     ) ||
     Object.values(recipe.spec.runtime.environment).some(
       (value) =>
         Buffer.byteLength(value, "utf8") > 4_096 ||
         value.includes("\0") ||
-        !HOST_LOCAL_SAFE_VALUE.test(value),
+        !HOST_LOCAL_SAFE_ENVIRONMENT_VALUE.test(value),
     )
   ) {
     return "host-local vLLM serving values must be bounded safe text";

test/managed-inference-catalog-compiler.test.ts

@@ -171,6 +171,7 @@ describe("managed inference YAML profile contract", () => {
       },
       serve: {
         arguments: expect.arrayContaining([
+          { name: "--gpu-memory-utilization", value: 0.65 },
           { name: "--tool-call-parser", value: "step3p5" },
           { name: "--reasoning-parser", value: "step3p5" },
           {