[Doc] cookbook: Add Nemotron 3.5 Super VL

docs/cookbook/autoregressive/NVIDIA/Nemotron3.5-Lightning.mdx

@@ -1,7 +1,6 @@
 ---
 title: Nemotron3.5-Lightning
 description: "Deploy NVIDIA Nemotron 3.5 Lightning with SGLang — NVFP4 serving with MTP, DFlash, and DSpark speculative decoding, reasoning, and tool calling."
-tag: NEW
 ---
 
 ## Deployment

docs/cookbook/autoregressive/NVIDIA/Nemotron3.5-Super.mdx

@@ -0,0 +1,92 @@
+---
+title: NVIDIA Nemotron 3.5 Super VL
+description: "Deploy Nemotron 3.5 Super VL 120B-A12B with SGLang on B200, H200, RTX PRO 6000, and DGX Spark, with NVFP4 or BF16 and optional MTP."
+tag: NEW
+---
+
+## Deployment
+
+The **Python** commands assume SGLang source at commit `2c7511f6f7cf1cdf9ae568ad0fb25f291edb10da`. The **Docker** commands use the planned public image tag `lmsysorg/sglang:nemotron3-5-super`; that tag is not published yet, so the Docker commands are not part of the GSM8K validation below. The command panel selects either `nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4` or `nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16`. Set `HF_TOKEN` in the panel's Env control if the checkpoint requires access.
+
+Choose **B200, H200, RTX PRO 6000, or DGX Spark** on one node. B200 offers NVFP4 and BF16. H200 offers BF16; NVFP4 is greyed out. RTX PRO 6000 and DGX Spark offer NVFP4; BF16 is greyed out. Choose `bfloat16` or `float32` in **Mamba SSM DType**; `bfloat16` is the default. **High Throughput** leaves MTP off. **Low Latency** adds EAGLE MTP with top-k 1, three steps, and four draft tokens. These names identify configuration choices; no throughput or latency comparison was measured. Both command tabs use `sglang serve`.
+
+import { Deployment } from "/src/snippets/_deployment.jsx";
+import { config } from "/src/snippets/configs/nvidia/nemotron-3.5-super-vl.jsx";
+import { benchmarks } from "/src/snippets/configs/nvidia/nemotron-3.5-super-vl-benchmarks.jsx";
+
+<Deployment config={config} benchmarks={benchmarks} />
+
+## Playground
+
+The Playground inherits the selected recipe and lets you toggle the reasoning and tool-call parsers. Parser changes are separate experiments.
+
+import { Playground } from "/src/snippets/_playground.jsx";
+
+<Playground config={config} />
+
+## 1. Model introduction
+
+**NVIDIA Nemotron 3.5 Super VL 120B-A12B** is a vision-language checkpoint in the Nemotron-H hybrid MoE family. The early-access NGC checkpoint configuration identifies `NemotronH_Omni_Reasoning_V3` and includes a vision encoder. The command panel uses the [NVFP4](https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4) and [BF16](https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16) Hugging Face IDs; the source runs used the [NGC model](https://registry.ngc.nvidia.com/orgs/0767305323357365/teams/super3-5-vl-ea-0911/models/nemotron-3.5-super-vl-120b-a12b-ea20261008). Check the checkpoint pages for access terms and model details.
+
+The validation used text-only GSM8K prompts. It does **not** establish image or video request correctness, long-context behavior, or tool-call accuracy.
+
+## 2. Configuration tips and validation
+
+- **B200, NVFP4 or BF16:** both strategies use one eight-GPU TP8 node with FP8 E4M3 KV cache and 16,384-token configured context. Select the Mamba state dtype in the wizard; SGLang selects the attention backend and memory limits by default.
+- **H200, BF16:** both strategies use one eight-GPU TP8 node with FA3 attention, Triton MoE, and `--disable-flashinfer-autotune`.
+- **RTX PRO 6000, NVFP4:** both strategies served at TP1 on GPU 1 of an eight-GPU G4 machine. The configured context was 262,144 tokens; only short prompts were tested.
+- **DGX Spark, NVFP4:** both strategies completed 300-question text-only GSM8K runs at TP1 with `bfloat16` Mamba state. The `float32` choice has no completed accuracy run.
+- **KV precision:** these recipes deliberately use `--kv-cache-dtype fp8_e4m3` to match the validated runs. KV precision can affect output quality; no BF16-KV comparison was measured.
+
+Server information confirmed `speculative_algorithm: null` for High Throughput and `EAGLE` for Low Latency. The runs used installed early-access NGC weights, not the Hugging Face checkpoint IDs in the command panel or the planned public Docker image.
+
+The Deployment panel’s Benchmark card shows text-only GSM8K accuracy for measured selections and a pending state for unmeasured selections. The existing green Verified status marks a recipe with a completed source-run. These results do not validate the public checkpoint IDs or planned Docker image.
+
+## 3. Reasoning output
+
+The base recipes enable `--reasoning-parser nemotron_3`. In the tested chat completions, the thinking trace appeared in `message.reasoning_content` and the final answer in `message.content`. The tool-call parser is configured as `qwen3_coder`, but structured tool calls were not part of this accuracy check.
+
+<Accordion title="Reasoning example (Python client)">
+
+```python Example
+from openai import OpenAI
+
+client = OpenAI(base_url="http://127.0.0.1:30000/v1", api_key="EMPTY")
+model = client.models.list().data[0].id
+response = client.chat.completions.create(
+    model=model,
+    messages=[{
+        "role": "user",
+        "content": "Claire makes a 3 egg omelet every morning for breakfast.  How many dozens of eggs will she eat in 4 weeks?\nSolve the problem and put the final numeric answer after ####.",
+    }],
+)
+message = response.choices[0].message
+print("Reasoning:", message.reasoning_content)
+print("Content:", message.content)
+```
+
+</Accordion>
+
+<Accordion title="Example output from the B200 NVFP4 smoke run">
+
+```text Output
+Reasoning: The user wants me to solve a math problem.
+
+Claire makes a 3 egg omelet every morning.
+She eats 3 eggs per day.
+4 weeks = 4 × 7 = 28 days.
+Total eggs = 3 × 28 = 84 eggs.
+Dozens of eggs = 84 / 12 = 7 dozens.
+
+Final answer: 7.
+
+I need to put the final numeric answer after ####.
+Content: Claire eats 3 eggs per day.
+In 4 weeks, there are 4 × 7 = 28 days.
+Total eggs eaten: 3 × 28 = 84 eggs.
+Since 1 dozen = 12 eggs: 84 ÷ 12 = 7 dozens.
+
+#### 7
+```
+
+</Accordion>

docs/cookbook/autoregressive/intro.mdx

@@ -106,7 +106,7 @@ metatags:
   <Card
     title="NVIDIA"
     mode="card"
-    href="/cookbook/autoregressive/NVIDIA/Nemotron3.5-Lightning"
+    href="/cookbook/autoregressive/NVIDIA/Nemotron3.5-Super"
     img="/cards/logos/nvidia.png"
   />
   <Card

docs/docs.json

@@ -1372,6 +1372,7 @@
                   {
                     "group": "NVIDIA",
                     "pages": [
+                      "cookbook/autoregressive/NVIDIA/Nemotron3.5-Super",
                       "cookbook/autoregressive/NVIDIA/Nemotron3.5-Lightning",
                       "cookbook/autoregressive/NVIDIA/Nemotron3-Ultra",
                       "cookbook/autoregressive/NVIDIA/Nemotron3-Nano-Omni",

docs/src/snippets/configs/nvidia/nemotron-3.5-super-vl-benchmarks.jsx

@@ -0,0 +1,77 @@
+// Text-only GSM8K scores from completed sgl-eval==0.1.2 runs.
+// Accuracy verification applies to these NGC-weight source runs, not to the
+// public Hugging Face checkpoint IDs or the planned cookbook Docker tag.
+export const benchmarks = [
+  {
+    match: { hw: "b200", variant: "default", quant: "nvfp4", strategy: "high-throughput", nodes: "single", mambaDtype: "bfloat16" },
+    sglang_version: "dev @ 2c7511f6f7cf",
+    accuracy: { gsm8k_pct: 95.98 },
+    notes: "1,266/1,319 correct; 20 outputs hit the 8,192-token cap; zero request errors. Text-only NGC checkpoint run.",
+  },
+  {
+    match: { hw: "b200", variant: "default", quant: "nvfp4", strategy: "high-throughput", nodes: "single", mambaDtype: "float32" },
+    sglang_version: "dev @ 2c7511f6f7cf",
+    accuracy: { gsm8k_pct: 97.33 },
+    notes: "292/300 correct on a smaller cohort; three outputs hit the 8,192-token cap; zero request errors. Text-only NGC checkpoint run.",
+  },
+  {
+    match: { hw: "b200", variant: "default", quant: "nvfp4", strategy: "low-latency", nodes: "single", mambaDtype: "bfloat16" },
+    sglang_version: "dev @ 2c7511f6f7cf",
+    accuracy: { gsm8k_pct: 96.13 },
+    notes: "1,268/1,319 correct; 16 outputs hit the 8,192-token cap; zero request errors. Text-only NGC checkpoint run.",
+  },
+  {
+    match: { hw: "b200", variant: "default", quant: "bf16", strategy: "high-throughput", nodes: "single", mambaDtype: "bfloat16" },
+    sglang_version: "dev @ 2c7511f6f7cf",
+    accuracy: { gsm8k_pct: 95.68 },
+    notes: "1,262/1,319 correct; 25 outputs hit the 8,192-token cap; zero request errors. Text-only NGC checkpoint run.",
+  },
+  {
+    match: { hw: "b200", variant: "default", quant: "bf16", strategy: "low-latency", nodes: "single", mambaDtype: "bfloat16" },
+    sglang_version: "dev @ 2c7511f6f7cf",
+    accuracy: { gsm8k_pct: 96.44 },
+    notes: "1,272/1,319 correct; 21 outputs hit the 8,192-token cap; zero request errors. Text-only NGC checkpoint run.",
+  },
+  {
+    match: { hw: "h200", variant: "default", quant: "bf16", strategy: "high-throughput", nodes: "single", mambaDtype: "bfloat16" },
+    sglang_version: "dev @ 2c7511f6f7cf",
+    accuracy: { gsm8k_pct: 96.06 },
+    notes: "1,267/1,319 correct; 21 outputs hit the 8,192-token cap; zero request errors. Text-only NGC checkpoint run.",
+  },
+  {
+    match: { hw: "h200", variant: "default", quant: "bf16", strategy: "low-latency", nodes: "single", mambaDtype: "bfloat16" },
+    sglang_version: "dev @ 2c7511f6f7cf",
+    accuracy: { gsm8k_pct: 95.83 },
+    notes: "1,264/1,319 correct; 21 outputs hit the 8,192-token cap; zero request errors. Text-only NGC checkpoint run.",
+  },
+  {
+    match: { hw: "rtx-pro-6000", variant: "default", quant: "nvfp4", strategy: "high-throughput", nodes: "single", mambaDtype: "bfloat16" },
+    sglang_version: "dev @ 2c7511f6f7cf",
+    accuracy: { gsm8k_pct: 96.06 },
+    notes: "1,267/1,319 correct; 21 outputs hit the 8,192-token cap; zero request errors. Text-only NGC checkpoint run.",
+  },
+  {
+    match: { hw: "rtx-pro-6000", variant: "default", quant: "nvfp4", strategy: "high-throughput", nodes: "single", mambaDtype: "float32" },
+    sglang_version: "dev @ 2c7511f6f7cf",
+    accuracy: { gsm8k_pct: 97.0 },
+    notes: "291/300 correct on a smaller cohort; five outputs hit the 8,192-token cap; zero request errors. Text-only NGC checkpoint run.",
+  },
+  {
+    match: { hw: "rtx-pro-6000", variant: "default", quant: "nvfp4", strategy: "low-latency", nodes: "single", mambaDtype: "bfloat16" },
+    sglang_version: "dev @ 2c7511f6f7cf",
+    accuracy: { gsm8k_pct: 96.44 },
+    notes: "1,272/1,319 correct; 18 outputs hit the 8,192-token cap; zero request errors. Text-only NGC checkpoint run.",
+  },
+  {
+    match: { hw: "dgx-spark", variant: "default", quant: "nvfp4", strategy: "high-throughput", nodes: "single", mambaDtype: "bfloat16" },
+    sglang_version: "dev @ 2c7511f6f7cf",
+    accuracy: { gsm8k_pct: 96.67 },
+    notes: "290/300 correct on a smaller cohort; five outputs hit the 8,192-token cap; zero request errors. Text-only NGC checkpoint run.",
+  },
+  {
+    match: { hw: "dgx-spark", variant: "default", quant: "nvfp4", strategy: "low-latency", nodes: "single", mambaDtype: "bfloat16" },
+    sglang_version: "dev @ 2c7511f6f7cf",
+    accuracy: { gsm8k_pct: 98.0 },
+    notes: "294/300 correct on a smaller cohort; three outputs hit the 8,192-token cap; zero request errors. Text-only NGC checkpoint run.",
+  },
+];