Update Nemotron 3 Nano Omni documentation and layout for Jetson Thor support

- Added build URL for Nemotron 3 Nano Omni.
- Updated minimum Jetson model from Orin AGX to Thor.
- Removed unsupported AGX Orin commands and references from the documentation.
- Enhanced layout logic to manage module commands based on active inference engine.
- Introduced SGLang support in the inference engines configuration.

Author: adsahu-nv

src/content/models/nemotron-3-nano-omni.md

@@ -17,7 +17,8 @@ license: "NVIDIA Open Model License"
 model_size: "21GB"
 hf_checkpoint: "nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-NVFP4"
 huggingface_url: "https://huggingface.co/nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-NVFP4"
-minimum_jetson: "Orin AGX"
+build_nvidia_url: "https://build.nvidia.com/nvidia/nemotron-3-nano-omni-30b-a3b-reasoning"
+minimum_jetson: "Thor"
 supported_inference_engines:
   - engine: "vLLM"
     type: "Container"
@@ -60,20 +61,6 @@ supported_inference_engines:
     modules_supported:
       - thor_t5000
       - thor_t4000
-      - orin_agx_64
-    serve_command_orin: |-
-      sudo docker run -it --rm --pull always \
-        --runtime=nvidia --network host \
-        -v $HOME/.cache/huggingface:/root/.cache/huggingface \
-        ghcr.io/nvidia-ai-iot/llama_cpp:latest-jetson-orin \
-        llama-server \
-          --hf-repo ggml-org/NVIDIA-Nemotron-3-Nano-Omni \
-          --hf-file nemotron-3-nano-omni-ga_v1.0-Q4_K_M.gguf \
-          --host 0.0.0.0 \
-          --port 8080 \
-          -c 8192 \
-          -ngl 999 \
-          --alias my_model
     serve_command_thor: |-
       sudo docker run -it --rm --pull always \
         --runtime=nvidia --network host \
@@ -110,8 +97,7 @@ Nemotron Nano 3 Omni is NVIDIA's multimodal reasoning model combining language,
 
 ## Supported Platforms
 
-- Jetson Thor (NVFP4, FP8, BF16 — vLLM, SGLang, llama.cpp)
-- Jetson Orin AGX 64GB (llama.cpp)
+- Jetson Thor  
 
 ## Running with vLLM
 
@@ -173,28 +159,9 @@ curl -s http://${JETSON_HOST}:30000/v1/chat/completions \
 
 <div class="device-tabs">
 <div class="device-tab-bar">
-<button class="device-tab active" data-target="orin-llama">Jetson Orin</button>
-<button class="device-tab" data-target="thor-llama">Jetson Thor</button>
-</div>
-<div class="device-panel" data-panel="orin-llama">
-
-```bash
-sudo docker run -it --rm --pull always \
-  --runtime=nvidia --network host \
-  -v $HOME/.cache/huggingface:/root/.cache/huggingface \
-  ghcr.io/nvidia-ai-iot/llama_cpp:latest-jetson-orin \
-  llama-server \
-    --hf-repo ggml-org/NVIDIA-Nemotron-3-Nano-Omni \
-    --hf-file nemotron-3-nano-omni-ga_v1.0-Q4_K_M.gguf \
-    --host 0.0.0.0 \
-    --port 8080 \
-    -c 8192 \
-    -ngl 999 \
-    --alias my_model
-```
-
+<button class="device-tab active" data-target="thor-llama">Jetson Thor</button>
 </div>
-<div class="device-panel hidden" data-panel="thor-llama">
+<div class="device-panel" data-panel="thor-llama">
 
 ```bash
 sudo docker run -it --rm --pull always \
@@ -231,8 +198,6 @@ curl -s http://127.0.0.1:8080/v1/chat/completions \
 
 > **Note:** `--hf-repo ggml-org/NVIDIA-Nemotron-3-Nano-Omni` and `--hf-file nemotron-3-nano-omni-ga_v1.0-Q4_K_M.gguf` download the official GGUF checkpoint from Hugging Face. `-ngl 999` offloads all layers to GPU. `--alias my_model` sets the model name used in API requests. `chat_template_kwargs: {"enable_thinking": true}` activates chain-of-thought reasoning.
 
--->
-
 ## Running with TensorRT Edge-LLM
 
 TensorRT Edge-LLM support for this model is currently Jetson Thor only. It requires exporting the model to ONNX and building TensorRT engines before running inference. See the [TensorRT Edge-LLM GitHub repository](https://github.com/NVIDIA/TensorRT-Edge-LLM) and the [export/build quick start](https://nvidia.github.io/TensorRT-Edge-LLM/user_guide/getting_started/quick-start-guide.html) for the full setup flow.

src/layouts/Layout.astro

@@ -741,9 +741,38 @@ const canonicalURL = new URL(Astro.url.pathname, Astro.site);
             syncRunModalMatrixEngineRail(modal, meta);
             return;
           }
+          var activeEngine = modal.getAttribute('data-active-engine') || meta.defaultEngineId || '';
+          var engineSupportMap = meta.modulesSupportedByEngine || {};
+          var allowedForActiveEngine = null;
+          if (Object.prototype.hasOwnProperty.call(engineSupportMap, activeEngine)) {
+            allowedForActiveEngine = engineSupportMap[activeEngine];
+          } else {
+            for (var ek in engineSupportMap) {
+              if (Object.prototype.hasOwnProperty.call(engineSupportMap, ek) && normalizeInferenceEngineKey(ek) === normalizeInferenceEngineKey(activeEngine)) {
+                allowedForActiveEngine = engineSupportMap[ek];
+                break;
+              }
+            }
+          }
+          function commandForModuleAndEngine(moduleId, engineName) {
+            var commands = meta.moduleCommandMatrix && meta.moduleCommandMatrix[moduleId];
+            if (!commands) return '';
+            for (var ck in commands) {
+              if (Object.prototype.hasOwnProperty.call(commands, ck) && normalizeInferenceEngineKey(ck) === normalizeInferenceEngineKey(engineName)) {
+                return commands[ck] || '';
+              }
+            }
+            return '';
+          }
           modal.querySelectorAll('[data-run-module-tab]').forEach(function (btn) {
             var global = btn.getAttribute('data-run-module-global-disabled') === '1';
-            var dis = global;
+            var moduleId = btn.getAttribute('data-run-module-tab') || '';
+            var unsupportedByEngine = false;
+            if (allowedForActiveEngine != null && Array.isArray(allowedForActiveEngine)) {
+              unsupportedByEngine = allowedForActiveEngine.length === 0 || allowedForActiveEngine.indexOf(moduleId) === -1;
+            }
+            var hasCmd = !!String(commandForModuleAndEngine(moduleId, activeEngine)).trim();
+            var dis = global || unsupportedByEngine || !hasCmd;
             btn.disabled = dis;
             if (dis) {
               applyMatrixModuleTabDisabledVisual(btn);

src/lib/engines.ts

@@ -33,6 +33,15 @@ export const INFERENCE_ENGINES: Record<string, InferenceEngine> = {
 			return `vllm serve ${checkpoint}`;
 		}
 	},
+	sglang: {
+		id: 'sglang',
+		label: 'SGLang',
+		supports: ['Text', 'Multimodal'],
+		buildCommand: ({ options }) => {
+			const checkpoint = options?.checkpoint || 'model/checkpoint';
+			return `sglang serve ${checkpoint}`;
+		}
+	},
 	edgemllm: {
 		id: 'edgemllm',
 		label: 'EdgeLLM',