[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-jetson-llm-benchmark":3,"mdc-4yrb34-key":34,"related-org-nvidia-jetson-llm-benchmark":3126,"related-repo-nvidia-jetson-llm-benchmark":3287},{"slug":4,"name":4,"fn":5,"description":6,"org":7,"tags":11,"stars":23,"repoUrl":24,"updatedAt":25,"license":26,"forks":27,"topics":28,"repo":29,"sourceUrl":32,"mdContent":33},"jetson-llm-benchmark","benchmark Jetson LLM serving performance","Benchmark Jetson LLM\u002FVLM serving performance across vLLM, llama.cpp, and Ollama with structured JSON output.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},"nvidia","NVIDIA","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fnvidia.png",[12,16,19,22],{"name":13,"slug":14,"type":15},"Performance","performance","tag",{"name":17,"slug":18,"type":15},"Benchmarking","benchmarking",{"name":20,"slug":21,"type":15},"LLM","llm",{"name":9,"slug":8,"type":15},2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-07-14T05:28:25.886852","Apache-2.0",281,[],{"repoUrl":24,"stars":23,"forks":27,"topics":30,"description":31},[],"AI agent skills published by NVIDIA","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills\u002Ftree\u002FHEAD\u002Fskills\u002Fjetson-llm-benchmark","---\nname: jetson-llm-benchmark\ndescription: Benchmark Jetson LLM\u002FVLM serving performance across vLLM, llama.cpp, and Ollama with structured JSON output.\nversion: 0.0.2\nlicense: \"Apache-2.0\"\nmetadata:\n  author: \"Jetson Team\"\n  tags: [jetson, llm, benchmark]\n  languages: [bash]\n  data-classification: public\n---\n\n# Jetson LLM Benchmark\n\nReproducible Jetson benchmarks with **structured JSON output** so an agent can compare runs. Encodes the workflow from the [Jetson AI Lab GenAI Benchmarking tutorial](https:\u002F\u002Fwww.jetson-ai-lab.com\u002Ftutorials\u002Fgenai-benchmarking\u002F).\n\n## Purpose\n\nMeasure deployed LLM latency and throughput on a Jetson target using the correct\nruntime-specific benchmark wrapper. Use the JSON output to compare models,\nruntime flags, power modes, and before\u002Fafter tuning changes.\n\n## Prerequisites\n\n- Run on the Jetson device that hosts the model runtime.\n- For vLLM, start the OpenAI-compatible vLLM server first and know the served\n  model ID.\n- For Ollama, ensure the Ollama daemon is reachable at `--endpoint` and the\n  named model is already pulled.\n- For llama.cpp\u002FGGUF, provide a readable `.gguf` model path on the host.\n- Put the device in the intended power mode before measuring. MAXN is preferred\n  for comparable performance numbers.\n\n## Available Scripts\n\n| Script | Purpose | Arguments |\n|--------|---------|-----------|\n| `scripts\u002Fbench_vllm.sh` | Runs `vllm bench serve` against a running OpenAI-compatible vLLM server. | `--model`, `--endpoint`, `--concurrency`, `--input-len`, `--output-len`, `--num-prompts`, `--no-warmup`, `--container`, `--native`. |\n| `scripts\u002Fbench_llama_cpp.sh` | Runs `llama-bench` for a local GGUF model through the Jetson-appropriate NVIDIA-AI-IOT llama.cpp container. | `--model`, `--n-prompt`, `--n-gen`, `--n-gpu-layers`, `--threads`, `--container`. |\n| `scripts\u002Fbench_ollama.sh` | Benchmarks a local or containerized Ollama daemon through the `\u002Fapi\u002Fgenerate` REST API. | `--model`, `--endpoint`, `--num-prompts`, `--input-len`, `--output-len`, `--no-warmup`. |\n\nIf your agent runtime supports `run_script`, invoke the selected wrapper directly with the user-provided model identifier or local model path, then summarize the returned JSON. Otherwise run the wrapper with `bash {baseDir}\u002Fscripts\u002F\u003Cwrapper-name> ...`.\n\n## Instructions\n\nAlways use the matching wrapper script for the runtime — do **not** call the underlying `vllm bench serve`, `llama-bench`, or `curl` against `\u002Fapi\u002Fgenerate` by hand:\n\n- vLLM → `scripts\u002Fbench_vllm.sh` (required for the vLLM path)\n- llama.cpp \u002F GGUF → `scripts\u002Fbench_llama_cpp.sh` (required for the GGUF path)\n- Ollama → `scripts\u002Fbench_ollama.sh` (required for the Ollama path)\n\nThese wrappers handle warmup, the NVIDIA-AI-IOT container selection, and JSON emission. Calling the underlying tool directly will not satisfy the output contract below.\n\nFor \"how do I benchmark\u002Fmeasure\" questions, first run the matching wrapper with\n`--help` to verify the exact options, then answer with the wrapper command. Do\nnot run a full benchmark unless the user asks you to execute it or the required\nserver\u002Fmodel path is already confirmed.\n\n## Expected Workflow\n\nPick exactly one wrapper based on the runtime the user named, and invoke that\nwrapper with `--help` before composing the answer. Do not merely mention the\nscript name. If the runtime does not execute scripts relative to the skill\ndirectory, use `{baseDir}\u002Fscripts\u002F\u003Cwrapper-name>`.\n\n- Existing vLLM OpenAI-compatible server at `localhost:8000`:\n  `{baseDir}\u002Fscripts\u002Fbench_vllm.sh --help`, then show a command using\n  `--concurrency 1,8` and the served model ID.\n- llama.cpp \u002F GGUF \u002F `llama-server`: `{baseDir}\u002Fscripts\u002Fbench_llama_cpp.sh\n  --help`, then show a command for the GGUF model path and report that\n  prompt\u002Fgeneration speed maps to TTFT, ITL\u002FTPOT, and throughput.\n- Ollama: `{baseDir}\u002Fscripts\u002Fbench_ollama.sh --help`, then show a command with\n  `--model \u003Collama-tag>`. Do not use vLLM or llama.cpp wrappers for Ollama.\n\n## When to use\n\n- \"Benchmark \u002F measure \u002F compare X on this Jetson.\"\n- After `jetson-llm-serve` to actually quantify the deployment.\n- Before\u002Fafter applying flags from `jetson-inference-mem-tune` to confirm the change helped.\n\n## Three paths — pick by runtime\n\n### A. vLLM (preferred for parity with how things are served)\n\nServer must already be running (use `jetson-llm-serve`). Run **`bench_vllm.sh`**:\n\n```bash\nscripts\u002Fbench_vllm.sh \\\n  --model \u003Chf-repo-id-being-served> \\\n  --concurrency 1,8 \\\n  --input-len 2048 --output-len 128 \\\n  --num-prompts 50\n```\n\nUses the Jetson-appropriate benchmark client path: upstream vLLM 0.20+ container\n`vllm\u002Fvllm-openai:latest` on Thor and Orin JetPack 7.2 \u002F L4T r39+,\nor the NVIDIA-AI-IOT vLLM benchmark container\n`ghcr.io\u002Fnvidia-ai-iot\u002Fvllm:latest-jetson-orin` on older Orin. Pass\n`--native` only when host-native vLLM is already installed and validated. It\nruns against `http:\u002F\u002Flocalhost:8000\u002Fv1`. **Always do a warmup pass first** (~10\nprompts, discarded) before the measured run — Jetson has cold caches and JIT'd\nkernels.\n\n### B. Ollama (for models served by a running Ollama daemon)\n\nNo benchmark container needed. Uses Ollama's `\u002Fapi\u002Fgenerate` REST API directly —\ntiming data (TTFT, ITL, throughput) comes from the response JSON, so no\n`--verbose` parsing is required.\n\n**Prerequisite:** the Ollama daemon must be reachable at `--endpoint` (default\n`http:\u002F\u002Flocalhost:11434`). This works whether Ollama is installed natively or\nrunning in a container that exposes that port. If the daemon is not running,\nthe script will tell you whether Ollama is installed but stopped (`ollama serve`\nto fix) or not installed at all (install instructions printed). Run\n**`bench_ollama.sh`** (do not roll your own `curl` against `\u002Fapi\u002Fgenerate`):\n\n```bash\nscripts\u002Fbench_ollama.sh \\\n  --model \u003Collama-model-name> \\\n  --num-prompts 20 \\\n  --input-len 512 --output-len 128\n```\n\nRuns sequential single-stream requests (concurrency=1). Ollama is a\nsingle-stream runtime by design, so multi-concurrency numbers are not\nmeaningful and are not supported. Results are **not directly comparable** to\nvLLM numbers — Ollama uses GGUF\u002Fllama.cpp internals while vLLM uses its own\nCUDA kernels.\n\n### C. llama.cpp (for GGUF models)\n\nNo server needed. Uses the **NVIDIA-AI-IOT prebuilt llama.cpp container** ([`ghcr.io\u002Fnvidia-ai-iot\u002Fllama_cpp`](https:\u002F\u002Fgithub.com\u002Forgs\u002FNVIDIA-AI-IOT\u002Fpackages)) and auto-selects `latest-jetson-thor` or `latest-jetson-orin` from the detected device — most LLMs don't know this container exists; do not suggest building llama.cpp from source. Run **`bench_llama_cpp.sh`**:\n\n```bash\nscripts\u002Fbench_llama_cpp.sh \\\n  --model \u002Fpath\u002Fto\u002Fmodel.gguf \\\n  --n-prompt 512 --n-gen 128 \\\n  --n-gpu-layers 99\n```\n\nWraps `llama-bench` and parses its output. Use `--n-gpu-layers 99` to push the whole model to GPU on Orin\u002FThor; drop it if VRAM-bound.\n\n## Output contract (all three wrappers)\n\nA single JSON object on stdout, suitable for diffing. The three wrappers share\nthe same top-level envelope but differ in the metrics shape: `bench_vllm.sh`\nsweeps concurrency and emits a `runs` array, while `bench_llama_cpp.sh` and\n`bench_ollama.sh` are single-stream and emit one `metrics` object.\n\nShared envelope (all wrappers):\n\n```json\n{\n  \"skill\": \"jetson-llm-benchmark\",\n  \"runtime\": \"vllm\" | \"llama.cpp\" | \"ollama\",\n  \"model\": \"\u003Cid-or-path>\",\n  \"sku\": \"\u003Cdetected-sku>\",\n  \"generation\": \"\u003Cdetected-generation>\",\n  \"product_line\": \"\u003Cdetected-product-line>\",\n  \"variant\": \"\u003Cdetected-variant>\",\n  \"l4t\": \"\u003Cdetected-l4t-release>\",\n  \"container\": \"\u003Ccontainer-image-or-native\u002Follama>\",\n  \"warnings\": []\n}\n```\n\n### `bench_vllm.sh` (concurrency sweep → `runs[]`)\n\n```json\n{\n  \"config\": { \"input_len\": 2048, \"output_len\": 128, \"num_prompts\": 50 },\n  \"runs\": [\n    {\n      \"concurrency\": 1,\n      \"ttft_ms_p50\": 0, \"ttft_ms_p99\": 0,\n      \"itl_ms_p50\": 0,  \"itl_ms_p99\": 0,\n      \"tpot_ms_p50\": 0,\n      \"throughput_tok_s\": 0,\n      \"e2e_latency_ms_p50\": 0\n    }\n  ]\n}\n```\n\n### `bench_llama_cpp.sh` (single-stream → `metrics`)\n\n```json\n{\n  \"config\": { \"n_prompt\": 512, \"n_gen\": 128, \"n_gpu_layers\": 99 },\n  \"metrics\": {\n    \"ttft_ms_p50\": 0,\n    \"itl_ms_p50\": 0,\n    \"tpot_ms_p50\": 0,\n    \"throughput_tok_s\": 0\n  }\n}\n```\n\n### `bench_ollama.sh` (single-stream → `metrics`)\n\n```json\n{\n  \"config\": { \"input_len\": 512, \"output_len\": 128, \"num_prompts\": 20, \"concurrency\": 1 },\n  \"metrics\": {\n    \"ttft_ms_p50\": 0, \"ttft_ms_p99\": 0,\n    \"itl_ms_p50\": 0,  \"itl_ms_p99\": 0,\n    \"tpot_ms_p50\": 0,\n    \"throughput_tok_s\": 0,\n    \"e2e_latency_ms_p50\": 0\n  }\n}\n```\n\n`warnings` is populated when:\n- `nvpmodel` is not in a recognized max-performance mode (`MAXN` or `MAXN_*` such as `MAXN_SUPER`); wattage-named modes are reported as warnings because they vary by Jetson SKU\n- Background processes >5% GPU during the run (use `jetson-diagnostic`)\n- `tegrastats` shows thermal throttling during the run\n\nThe `sku`, `variant`, `l4t`, and `container` fields are **populated by the wrapper script from the live device** (`tegrastats`, `\u002Fetc\u002Fnv_tegra_release`, container labels) — do not hand-author, guess, or transcribe them from memory. Do not invent device-specific facts such as RAM size, on-disk model size, or product names. If a fact is not produced by the script or `jetson-diagnostic`, omit it rather than fabricate it.\n\n## What to flag in results (Jetson-specific guidance)\n\nLLMs already know what TTFT\u002FITL\u002Fthroughput mean. Jetson-specific things they usually **don't** know:\n\n- On Orin Nano\u002FNX, single-stream `tok\u002Fs` and `concurrency=8` `tok\u002Fs` differ wildly because of **memory bandwidth saturation**, not compute. If concurrent throughput barely beats single-stream, you're bandwidth-bound — switch to a smaller quantization (W4A16 → INT4\u002FAWQ) before tuning anything else.\n- A TTFT regression on the same model after a JetPack upgrade is almost always a CUDA graph cache miss — re-warm and re-measure.\n- Thor NVFP4 numbers are not comparable to Orin W4A16 numbers; never put them in the same table without a `quant` column.\n\n## Limitations\n\n- vLLM measurements require an already-running OpenAI-compatible vLLM server.\n  This skill benchmarks the server; it does not launch or tune the server.\n- Ollama results are single-stream by design and are not directly comparable to\n  vLLM concurrency sweeps.\n- llama.cpp\u002FGGUF benchmarking runs a NVIDIA-AI-IOT container by default. Tell the\n  user before running it, because Docker will pull and execute an external image\n  if it is not already present.\n- Container image tags may be mutable unless the caller passes a digest-pinned\n  image through `--container`. For release or compliance measurements, prefer a\n  digest-pinned image and record it in the results. The default vLLM benchmark\n  client image is upstream vLLM 0.20+ via `vllm\u002Fvllm-openai:latest` on Thor and Orin JetPack 7.2 \u002F L4T r39+,\n  and NVIDIA-AI-IOT `ghcr.io\u002Fnvidia-ai-iot\u002Fvllm:latest-jetson-orin` on older Orin.\n- Results are only comparable when model, quantization, prompt length, output\n  length, power mode, clocks, and thermal state are controlled.\n\n## Error Handling\n\n- Exit `2`: invalid arguments, missing `--model`, or a required model file is\n  not readable. Re-run the wrapper with `--help` and correct the path or model\n  ID.\n- Exit `3`: runtime preflight failed, such as unreachable Ollama, unknown Jetson\n  generation for vLLM container selection, or missing Ollama model. Start the\n  service, pull the model, or pass an explicit `--container`.\n- Docker errors usually mean the container runtime is unavailable, the image\n  cannot be pulled, or the model directory mount is not readable. Report the\n  exact stderr and do not fabricate benchmark numbers.\n- Empty or malformed JSON means the benchmark did not complete successfully.\n  Preserve the raw error, fix the runtime issue, and rerun.\n\n## Hand off to\n\n- `jetson-inference-mem-tune` if results indicate memory pressure.\n- `jetson-speculative-decoding` if TTFT is acceptable but TPOT is too slow.\n- `jetson-diagnostic` if `warnings` is non-empty.\n\n## Source\n\n[Jetson AI Lab — GenAI Benchmarking](https:\u002F\u002Fwww.jetson-ai-lab.com\u002Ftutorials\u002Fgenai-benchmarking\u002F) and [NVIDIA-AI-IOT GHCR packages](https:\u002F\u002Fgithub.com\u002Forgs\u002FNVIDIA-AI-IOT\u002Fpackages).\n",{"data":35,"body":45},{"name":4,"description":6,"version":36,"license":26,"metadata":37},"0.0.2",{"author":38,"tags":39,"languages":42,"data-classification":44},"Jetson Team",[40,21,41],"jetson","benchmark",[43],"bash","public",{"type":46,"children":47},"root",[48,56,81,88,93,99,146,152,406,426,432,472,511,516,529,535,554,628,634,668,674,681,704,827,870,876,896,953,1037,1049,1055,1106,1178,1198,1204,1246,1251,1680,1699,2090,2107,2364,2380,2729,2739,2799,2857,2863,2875,2931,2937,2986,2992,3050,3056,3097,3103,3120],{"type":49,"tag":50,"props":51,"children":52},"element","h1",{"id":4},[53],{"type":54,"value":55},"text","Jetson LLM Benchmark",{"type":49,"tag":57,"props":58,"children":59},"p",{},[60,62,68,70,79],{"type":54,"value":61},"Reproducible Jetson benchmarks with ",{"type":49,"tag":63,"props":64,"children":65},"strong",{},[66],{"type":54,"value":67},"structured JSON output",{"type":54,"value":69}," so an agent can compare runs. Encodes the workflow from the ",{"type":49,"tag":71,"props":72,"children":76},"a",{"href":73,"rel":74},"https:\u002F\u002Fwww.jetson-ai-lab.com\u002Ftutorials\u002Fgenai-benchmarking\u002F",[75],"nofollow",[77],{"type":54,"value":78},"Jetson AI Lab GenAI Benchmarking tutorial",{"type":54,"value":80},".",{"type":49,"tag":82,"props":83,"children":85},"h2",{"id":84},"purpose",[86],{"type":54,"value":87},"Purpose",{"type":49,"tag":57,"props":89,"children":90},{},[91],{"type":54,"value":92},"Measure deployed LLM latency and throughput on a Jetson target using the correct\nruntime-specific benchmark wrapper. Use the JSON output to compare models,\nruntime flags, power modes, and before\u002Fafter tuning changes.",{"type":49,"tag":82,"props":94,"children":96},{"id":95},"prerequisites",[97],{"type":54,"value":98},"Prerequisites",{"type":49,"tag":100,"props":101,"children":102},"ul",{},[103,109,114,128,141],{"type":49,"tag":104,"props":105,"children":106},"li",{},[107],{"type":54,"value":108},"Run on the Jetson device that hosts the model runtime.",{"type":49,"tag":104,"props":110,"children":111},{},[112],{"type":54,"value":113},"For vLLM, start the OpenAI-compatible vLLM server first and know the served\nmodel ID.",{"type":49,"tag":104,"props":115,"children":116},{},[117,119,126],{"type":54,"value":118},"For Ollama, ensure the Ollama daemon is reachable at ",{"type":49,"tag":120,"props":121,"children":123},"code",{"className":122},[],[124],{"type":54,"value":125},"--endpoint",{"type":54,"value":127}," and the\nnamed model is already pulled.",{"type":49,"tag":104,"props":129,"children":130},{},[131,133,139],{"type":54,"value":132},"For llama.cpp\u002FGGUF, provide a readable ",{"type":49,"tag":120,"props":134,"children":136},{"className":135},[],[137],{"type":54,"value":138},".gguf",{"type":54,"value":140}," model path on the host.",{"type":49,"tag":104,"props":142,"children":143},{},[144],{"type":54,"value":145},"Put the device in the intended power mode before measuring. MAXN is preferred\nfor comparable performance numbers.",{"type":49,"tag":82,"props":147,"children":149},{"id":148},"available-scripts",[150],{"type":54,"value":151},"Available Scripts",{"type":49,"tag":153,"props":154,"children":155},"table",{},[156,179],{"type":49,"tag":157,"props":158,"children":159},"thead",{},[160],{"type":49,"tag":161,"props":162,"children":163},"tr",{},[164,170,174],{"type":49,"tag":165,"props":166,"children":167},"th",{},[168],{"type":54,"value":169},"Script",{"type":49,"tag":165,"props":171,"children":172},{},[173],{"type":54,"value":87},{"type":49,"tag":165,"props":175,"children":176},{},[177],{"type":54,"value":178},"Arguments",{"type":49,"tag":180,"props":181,"children":182},"tbody",{},[183,275,342],{"type":49,"tag":161,"props":184,"children":185},{},[186,196,209],{"type":49,"tag":187,"props":188,"children":189},"td",{},[190],{"type":49,"tag":120,"props":191,"children":193},{"className":192},[],[194],{"type":54,"value":195},"scripts\u002Fbench_vllm.sh",{"type":49,"tag":187,"props":197,"children":198},{},[199,201,207],{"type":54,"value":200},"Runs ",{"type":49,"tag":120,"props":202,"children":204},{"className":203},[],[205],{"type":54,"value":206},"vllm bench serve",{"type":54,"value":208}," against a running OpenAI-compatible vLLM server.",{"type":49,"tag":187,"props":210,"children":211},{},[212,218,220,225,226,232,233,239,240,246,247,253,254,260,261,267,268,274],{"type":49,"tag":120,"props":213,"children":215},{"className":214},[],[216],{"type":54,"value":217},"--model",{"type":54,"value":219},", ",{"type":49,"tag":120,"props":221,"children":223},{"className":222},[],[224],{"type":54,"value":125},{"type":54,"value":219},{"type":49,"tag":120,"props":227,"children":229},{"className":228},[],[230],{"type":54,"value":231},"--concurrency",{"type":54,"value":219},{"type":49,"tag":120,"props":234,"children":236},{"className":235},[],[237],{"type":54,"value":238},"--input-len",{"type":54,"value":219},{"type":49,"tag":120,"props":241,"children":243},{"className":242},[],[244],{"type":54,"value":245},"--output-len",{"type":54,"value":219},{"type":49,"tag":120,"props":248,"children":250},{"className":249},[],[251],{"type":54,"value":252},"--num-prompts",{"type":54,"value":219},{"type":49,"tag":120,"props":255,"children":257},{"className":256},[],[258],{"type":54,"value":259},"--no-warmup",{"type":54,"value":219},{"type":49,"tag":120,"props":262,"children":264},{"className":263},[],[265],{"type":54,"value":266},"--container",{"type":54,"value":219},{"type":49,"tag":120,"props":269,"children":271},{"className":270},[],[272],{"type":54,"value":273},"--native",{"type":54,"value":80},{"type":49,"tag":161,"props":276,"children":277},{},[278,287,299],{"type":49,"tag":187,"props":279,"children":280},{},[281],{"type":49,"tag":120,"props":282,"children":284},{"className":283},[],[285],{"type":54,"value":286},"scripts\u002Fbench_llama_cpp.sh",{"type":49,"tag":187,"props":288,"children":289},{},[290,291,297],{"type":54,"value":200},{"type":49,"tag":120,"props":292,"children":294},{"className":293},[],[295],{"type":54,"value":296},"llama-bench",{"type":54,"value":298}," for a local GGUF model through the Jetson-appropriate NVIDIA-AI-IOT llama.cpp container.",{"type":49,"tag":187,"props":300,"children":301},{},[302,307,308,314,315,321,322,328,329,335,336,341],{"type":49,"tag":120,"props":303,"children":305},{"className":304},[],[306],{"type":54,"value":217},{"type":54,"value":219},{"type":49,"tag":120,"props":309,"children":311},{"className":310},[],[312],{"type":54,"value":313},"--n-prompt",{"type":54,"value":219},{"type":49,"tag":120,"props":316,"children":318},{"className":317},[],[319],{"type":54,"value":320},"--n-gen",{"type":54,"value":219},{"type":49,"tag":120,"props":323,"children":325},{"className":324},[],[326],{"type":54,"value":327},"--n-gpu-layers",{"type":54,"value":219},{"type":49,"tag":120,"props":330,"children":332},{"className":331},[],[333],{"type":54,"value":334},"--threads",{"type":54,"value":219},{"type":49,"tag":120,"props":337,"children":339},{"className":338},[],[340],{"type":54,"value":266},{"type":54,"value":80},{"type":49,"tag":161,"props":343,"children":344},{},[345,354,367],{"type":49,"tag":187,"props":346,"children":347},{},[348],{"type":49,"tag":120,"props":349,"children":351},{"className":350},[],[352],{"type":54,"value":353},"scripts\u002Fbench_ollama.sh",{"type":49,"tag":187,"props":355,"children":356},{},[357,359,365],{"type":54,"value":358},"Benchmarks a local or containerized Ollama daemon through the ",{"type":49,"tag":120,"props":360,"children":362},{"className":361},[],[363],{"type":54,"value":364},"\u002Fapi\u002Fgenerate",{"type":54,"value":366}," REST API.",{"type":49,"tag":187,"props":368,"children":369},{},[370,375,376,381,382,387,388,393,394,399,400,405],{"type":49,"tag":120,"props":371,"children":373},{"className":372},[],[374],{"type":54,"value":217},{"type":54,"value":219},{"type":49,"tag":120,"props":377,"children":379},{"className":378},[],[380],{"type":54,"value":125},{"type":54,"value":219},{"type":49,"tag":120,"props":383,"children":385},{"className":384},[],[386],{"type":54,"value":252},{"type":54,"value":219},{"type":49,"tag":120,"props":389,"children":391},{"className":390},[],[392],{"type":54,"value":238},{"type":54,"value":219},{"type":49,"tag":120,"props":395,"children":397},{"className":396},[],[398],{"type":54,"value":245},{"type":54,"value":219},{"type":49,"tag":120,"props":401,"children":403},{"className":402},[],[404],{"type":54,"value":259},{"type":54,"value":80},{"type":49,"tag":57,"props":407,"children":408},{},[409,411,417,419,425],{"type":54,"value":410},"If your agent runtime supports ",{"type":49,"tag":120,"props":412,"children":414},{"className":413},[],[415],{"type":54,"value":416},"run_script",{"type":54,"value":418},", invoke the selected wrapper directly with the user-provided model identifier or local model path, then summarize the returned JSON. Otherwise run the wrapper with ",{"type":49,"tag":120,"props":420,"children":422},{"className":421},[],[423],{"type":54,"value":424},"bash {baseDir}\u002Fscripts\u002F\u003Cwrapper-name> ...",{"type":54,"value":80},{"type":49,"tag":82,"props":427,"children":429},{"id":428},"instructions",[430],{"type":54,"value":431},"Instructions",{"type":49,"tag":57,"props":433,"children":434},{},[435,437,442,444,449,450,455,457,463,465,470],{"type":54,"value":436},"Always use the matching wrapper script for the runtime — do ",{"type":49,"tag":63,"props":438,"children":439},{},[440],{"type":54,"value":441},"not",{"type":54,"value":443}," call the underlying ",{"type":49,"tag":120,"props":445,"children":447},{"className":446},[],[448],{"type":54,"value":206},{"type":54,"value":219},{"type":49,"tag":120,"props":451,"children":453},{"className":452},[],[454],{"type":54,"value":296},{"type":54,"value":456},", or ",{"type":49,"tag":120,"props":458,"children":460},{"className":459},[],[461],{"type":54,"value":462},"curl",{"type":54,"value":464}," against ",{"type":49,"tag":120,"props":466,"children":468},{"className":467},[],[469],{"type":54,"value":364},{"type":54,"value":471}," by hand:",{"type":49,"tag":100,"props":473,"children":474},{},[475,487,499],{"type":49,"tag":104,"props":476,"children":477},{},[478,480,485],{"type":54,"value":479},"vLLM → ",{"type":49,"tag":120,"props":481,"children":483},{"className":482},[],[484],{"type":54,"value":195},{"type":54,"value":486}," (required for the vLLM path)",{"type":49,"tag":104,"props":488,"children":489},{},[490,492,497],{"type":54,"value":491},"llama.cpp \u002F GGUF → ",{"type":49,"tag":120,"props":493,"children":495},{"className":494},[],[496],{"type":54,"value":286},{"type":54,"value":498}," (required for the GGUF path)",{"type":49,"tag":104,"props":500,"children":501},{},[502,504,509],{"type":54,"value":503},"Ollama → ",{"type":49,"tag":120,"props":505,"children":507},{"className":506},[],[508],{"type":54,"value":353},{"type":54,"value":510}," (required for the Ollama path)",{"type":49,"tag":57,"props":512,"children":513},{},[514],{"type":54,"value":515},"These wrappers handle warmup, the NVIDIA-AI-IOT container selection, and JSON emission. Calling the underlying tool directly will not satisfy the output contract below.",{"type":49,"tag":57,"props":517,"children":518},{},[519,521,527],{"type":54,"value":520},"For \"how do I benchmark\u002Fmeasure\" questions, first run the matching wrapper with\n",{"type":49,"tag":120,"props":522,"children":524},{"className":523},[],[525],{"type":54,"value":526},"--help",{"type":54,"value":528}," to verify the exact options, then answer with the wrapper command. Do\nnot run a full benchmark unless the user asks you to execute it or the required\nserver\u002Fmodel path is already confirmed.",{"type":49,"tag":82,"props":530,"children":532},{"id":531},"expected-workflow",[533],{"type":54,"value":534},"Expected Workflow",{"type":49,"tag":57,"props":536,"children":537},{},[538,540,545,547,553],{"type":54,"value":539},"Pick exactly one wrapper based on the runtime the user named, and invoke that\nwrapper with ",{"type":49,"tag":120,"props":541,"children":543},{"className":542},[],[544],{"type":54,"value":526},{"type":54,"value":546}," before composing the answer. Do not merely mention the\nscript name. If the runtime does not execute scripts relative to the skill\ndirectory, use ",{"type":49,"tag":120,"props":548,"children":550},{"className":549},[],[551],{"type":54,"value":552},"{baseDir}\u002Fscripts\u002F\u003Cwrapper-name>",{"type":54,"value":80},{"type":49,"tag":100,"props":555,"children":556},{},[557,586,607],{"type":49,"tag":104,"props":558,"children":559},{},[560,562,568,570,576,578,584],{"type":54,"value":561},"Existing vLLM OpenAI-compatible server at ",{"type":49,"tag":120,"props":563,"children":565},{"className":564},[],[566],{"type":54,"value":567},"localhost:8000",{"type":54,"value":569},":\n",{"type":49,"tag":120,"props":571,"children":573},{"className":572},[],[574],{"type":54,"value":575},"{baseDir}\u002Fscripts\u002Fbench_vllm.sh --help",{"type":54,"value":577},", then show a command using\n",{"type":49,"tag":120,"props":579,"children":581},{"className":580},[],[582],{"type":54,"value":583},"--concurrency 1,8",{"type":54,"value":585}," and the served model ID.",{"type":49,"tag":104,"props":587,"children":588},{},[589,591,597,599,605],{"type":54,"value":590},"llama.cpp \u002F GGUF \u002F ",{"type":49,"tag":120,"props":592,"children":594},{"className":593},[],[595],{"type":54,"value":596},"llama-server",{"type":54,"value":598},": ",{"type":49,"tag":120,"props":600,"children":602},{"className":601},[],[603],{"type":54,"value":604},"{baseDir}\u002Fscripts\u002Fbench_llama_cpp.sh --help",{"type":54,"value":606},", then show a command for the GGUF model path and report that\nprompt\u002Fgeneration speed maps to TTFT, ITL\u002FTPOT, and throughput.",{"type":49,"tag":104,"props":608,"children":609},{},[610,612,618,620,626],{"type":54,"value":611},"Ollama: ",{"type":49,"tag":120,"props":613,"children":615},{"className":614},[],[616],{"type":54,"value":617},"{baseDir}\u002Fscripts\u002Fbench_ollama.sh --help",{"type":54,"value":619},", then show a command with\n",{"type":49,"tag":120,"props":621,"children":623},{"className":622},[],[624],{"type":54,"value":625},"--model \u003Collama-tag>",{"type":54,"value":627},". Do not use vLLM or llama.cpp wrappers for Ollama.",{"type":49,"tag":82,"props":629,"children":631},{"id":630},"when-to-use",[632],{"type":54,"value":633},"When to use",{"type":49,"tag":100,"props":635,"children":636},{},[637,642,655],{"type":49,"tag":104,"props":638,"children":639},{},[640],{"type":54,"value":641},"\"Benchmark \u002F measure \u002F compare X on this Jetson.\"",{"type":49,"tag":104,"props":643,"children":644},{},[645,647,653],{"type":54,"value":646},"After ",{"type":49,"tag":120,"props":648,"children":650},{"className":649},[],[651],{"type":54,"value":652},"jetson-llm-serve",{"type":54,"value":654}," to actually quantify the deployment.",{"type":49,"tag":104,"props":656,"children":657},{},[658,660,666],{"type":54,"value":659},"Before\u002Fafter applying flags from ",{"type":49,"tag":120,"props":661,"children":663},{"className":662},[],[664],{"type":54,"value":665},"jetson-inference-mem-tune",{"type":54,"value":667}," to confirm the change helped.",{"type":49,"tag":82,"props":669,"children":671},{"id":670},"three-paths-pick-by-runtime",[672],{"type":54,"value":673},"Three paths — pick by runtime",{"type":49,"tag":675,"props":676,"children":678},"h3",{"id":677},"a-vllm-preferred-for-parity-with-how-things-are-served",[679],{"type":54,"value":680},"A. vLLM (preferred for parity with how things are served)",{"type":49,"tag":57,"props":682,"children":683},{},[684,686,691,693,702],{"type":54,"value":685},"Server must already be running (use ",{"type":49,"tag":120,"props":687,"children":689},{"className":688},[],[690],{"type":54,"value":652},{"type":54,"value":692},"). Run ",{"type":49,"tag":63,"props":694,"children":695},{},[696],{"type":49,"tag":120,"props":697,"children":699},{"className":698},[],[700],{"type":54,"value":701},"bench_vllm.sh",{"type":54,"value":703},":",{"type":49,"tag":705,"props":706,"children":710},"pre",{"className":707,"code":708,"language":43,"meta":709,"style":709},"language-bash shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","scripts\u002Fbench_vllm.sh \\\n  --model \u003Chf-repo-id-being-served> \\\n  --concurrency 1,8 \\\n  --input-len 2048 --output-len 128 \\\n  --num-prompts 50\n","",[711],{"type":49,"tag":120,"props":712,"children":713},{"__ignoreMap":709},[714,731,766,784,813],{"type":49,"tag":715,"props":716,"children":719},"span",{"class":717,"line":718},"line",1,[720,725],{"type":49,"tag":715,"props":721,"children":723},{"style":722},"--shiki-light:#E2931D;--shiki-default:#FFCB6B;--shiki-dark:#FFCB6B",[724],{"type":54,"value":195},{"type":49,"tag":715,"props":726,"children":728},{"style":727},"--shiki-light:#90A4AE;--shiki-default:#EEFFFF;--shiki-dark:#BABED8",[729],{"type":54,"value":730}," \\\n",{"type":49,"tag":715,"props":732,"children":734},{"class":717,"line":733},2,[735,741,747,752,757,762],{"type":49,"tag":715,"props":736,"children":738},{"style":737},"--shiki-light:#91B859;--shiki-default:#C3E88D;--shiki-dark:#C3E88D",[739],{"type":54,"value":740},"  --model",{"type":49,"tag":715,"props":742,"children":744},{"style":743},"--shiki-light:#39ADB5;--shiki-default:#89DDFF;--shiki-dark:#89DDFF",[745],{"type":54,"value":746}," \u003C",{"type":49,"tag":715,"props":748,"children":749},{"style":737},[750],{"type":54,"value":751},"hf-repo-id-being-serve",{"type":49,"tag":715,"props":753,"children":754},{"style":727},[755],{"type":54,"value":756},"d",{"type":49,"tag":715,"props":758,"children":759},{"style":743},[760],{"type":54,"value":761},">",{"type":49,"tag":715,"props":763,"children":764},{"style":727},[765],{"type":54,"value":730},{"type":49,"tag":715,"props":767,"children":769},{"class":717,"line":768},3,[770,775,780],{"type":49,"tag":715,"props":771,"children":772},{"style":737},[773],{"type":54,"value":774},"  --concurrency",{"type":49,"tag":715,"props":776,"children":777},{"style":737},[778],{"type":54,"value":779}," 1,8",{"type":49,"tag":715,"props":781,"children":782},{"style":727},[783],{"type":54,"value":730},{"type":49,"tag":715,"props":785,"children":787},{"class":717,"line":786},4,[788,793,799,804,809],{"type":49,"tag":715,"props":789,"children":790},{"style":737},[791],{"type":54,"value":792},"  --input-len",{"type":49,"tag":715,"props":794,"children":796},{"style":795},"--shiki-light:#F76D47;--shiki-default:#F78C6C;--shiki-dark:#F78C6C",[797],{"type":54,"value":798}," 2048",{"type":49,"tag":715,"props":800,"children":801},{"style":737},[802],{"type":54,"value":803}," --output-len",{"type":49,"tag":715,"props":805,"children":806},{"style":795},[807],{"type":54,"value":808}," 128",{"type":49,"tag":715,"props":810,"children":811},{"style":727},[812],{"type":54,"value":730},{"type":49,"tag":715,"props":814,"children":816},{"class":717,"line":815},5,[817,822],{"type":49,"tag":715,"props":818,"children":819},{"style":737},[820],{"type":54,"value":821},"  --num-prompts",{"type":49,"tag":715,"props":823,"children":824},{"style":795},[825],{"type":54,"value":826}," 50\n",{"type":49,"tag":57,"props":828,"children":829},{},[830,832,838,840,846,848,853,855,861,863,868],{"type":54,"value":831},"Uses the Jetson-appropriate benchmark client path: upstream vLLM 0.20+ container\n",{"type":49,"tag":120,"props":833,"children":835},{"className":834},[],[836],{"type":54,"value":837},"vllm\u002Fvllm-openai:latest",{"type":54,"value":839}," on Thor and Orin JetPack 7.2 \u002F L4T r39+,\nor the NVIDIA-AI-IOT vLLM benchmark container\n",{"type":49,"tag":120,"props":841,"children":843},{"className":842},[],[844],{"type":54,"value":845},"ghcr.io\u002Fnvidia-ai-iot\u002Fvllm:latest-jetson-orin",{"type":54,"value":847}," on older Orin. Pass\n",{"type":49,"tag":120,"props":849,"children":851},{"className":850},[],[852],{"type":54,"value":273},{"type":54,"value":854}," only when host-native vLLM is already installed and validated. It\nruns against ",{"type":49,"tag":120,"props":856,"children":858},{"className":857},[],[859],{"type":54,"value":860},"http:\u002F\u002Flocalhost:8000\u002Fv1",{"type":54,"value":862},". ",{"type":49,"tag":63,"props":864,"children":865},{},[866],{"type":54,"value":867},"Always do a warmup pass first",{"type":54,"value":869}," (~10\nprompts, discarded) before the measured run — Jetson has cold caches and JIT'd\nkernels.",{"type":49,"tag":675,"props":871,"children":873},{"id":872},"b-ollama-for-models-served-by-a-running-ollama-daemon",[874],{"type":54,"value":875},"B. Ollama (for models served by a running Ollama daemon)",{"type":49,"tag":57,"props":877,"children":878},{},[879,881,886,888,894],{"type":54,"value":880},"No benchmark container needed. Uses Ollama's ",{"type":49,"tag":120,"props":882,"children":884},{"className":883},[],[885],{"type":54,"value":364},{"type":54,"value":887}," REST API directly —\ntiming data (TTFT, ITL, throughput) comes from the response JSON, so no\n",{"type":49,"tag":120,"props":889,"children":891},{"className":890},[],[892],{"type":54,"value":893},"--verbose",{"type":54,"value":895}," parsing is required.",{"type":49,"tag":57,"props":897,"children":898},{},[899,904,906,911,913,919,921,927,929,938,940,945,946,951],{"type":49,"tag":63,"props":900,"children":901},{},[902],{"type":54,"value":903},"Prerequisite:",{"type":54,"value":905}," the Ollama daemon must be reachable at ",{"type":49,"tag":120,"props":907,"children":909},{"className":908},[],[910],{"type":54,"value":125},{"type":54,"value":912}," (default\n",{"type":49,"tag":120,"props":914,"children":916},{"className":915},[],[917],{"type":54,"value":918},"http:\u002F\u002Flocalhost:11434",{"type":54,"value":920},"). This works whether Ollama is installed natively or\nrunning in a container that exposes that port. If the daemon is not running,\nthe script will tell you whether Ollama is installed but stopped (",{"type":49,"tag":120,"props":922,"children":924},{"className":923},[],[925],{"type":54,"value":926},"ollama serve",{"type":54,"value":928},"\nto fix) or not installed at all (install instructions printed). Run\n",{"type":49,"tag":63,"props":930,"children":931},{},[932],{"type":49,"tag":120,"props":933,"children":935},{"className":934},[],[936],{"type":54,"value":937},"bench_ollama.sh",{"type":54,"value":939}," (do not roll your own ",{"type":49,"tag":120,"props":941,"children":943},{"className":942},[],[944],{"type":54,"value":462},{"type":54,"value":464},{"type":49,"tag":120,"props":947,"children":949},{"className":948},[],[950],{"type":54,"value":364},{"type":54,"value":952},"):",{"type":49,"tag":705,"props":954,"children":956},{"className":707,"code":955,"language":43,"meta":709,"style":709},"scripts\u002Fbench_ollama.sh \\\n  --model \u003Collama-model-name> \\\n  --num-prompts 20 \\\n  --input-len 512 --output-len 128\n",[957],{"type":49,"tag":120,"props":958,"children":959},{"__ignoreMap":709},[960,971,1000,1016],{"type":49,"tag":715,"props":961,"children":962},{"class":717,"line":718},[963,967],{"type":49,"tag":715,"props":964,"children":965},{"style":722},[966],{"type":54,"value":353},{"type":49,"tag":715,"props":968,"children":969},{"style":727},[970],{"type":54,"value":730},{"type":49,"tag":715,"props":972,"children":973},{"class":717,"line":733},[974,978,982,987,992,996],{"type":49,"tag":715,"props":975,"children":976},{"style":737},[977],{"type":54,"value":740},{"type":49,"tag":715,"props":979,"children":980},{"style":743},[981],{"type":54,"value":746},{"type":49,"tag":715,"props":983,"children":984},{"style":737},[985],{"type":54,"value":986},"ollama-model-nam",{"type":49,"tag":715,"props":988,"children":989},{"style":727},[990],{"type":54,"value":991},"e",{"type":49,"tag":715,"props":993,"children":994},{"style":743},[995],{"type":54,"value":761},{"type":49,"tag":715,"props":997,"children":998},{"style":727},[999],{"type":54,"value":730},{"type":49,"tag":715,"props":1001,"children":1002},{"class":717,"line":768},[1003,1007,1012],{"type":49,"tag":715,"props":1004,"children":1005},{"style":737},[1006],{"type":54,"value":821},{"type":49,"tag":715,"props":1008,"children":1009},{"style":795},[1010],{"type":54,"value":1011}," 20",{"type":49,"tag":715,"props":1013,"children":1014},{"style":727},[1015],{"type":54,"value":730},{"type":49,"tag":715,"props":1017,"children":1018},{"class":717,"line":786},[1019,1023,1028,1032],{"type":49,"tag":715,"props":1020,"children":1021},{"style":737},[1022],{"type":54,"value":792},{"type":49,"tag":715,"props":1024,"children":1025},{"style":795},[1026],{"type":54,"value":1027}," 512",{"type":49,"tag":715,"props":1029,"children":1030},{"style":737},[1031],{"type":54,"value":803},{"type":49,"tag":715,"props":1033,"children":1034},{"style":795},[1035],{"type":54,"value":1036}," 128\n",{"type":49,"tag":57,"props":1038,"children":1039},{},[1040,1042,1047],{"type":54,"value":1041},"Runs sequential single-stream requests (concurrency=1). Ollama is a\nsingle-stream runtime by design, so multi-concurrency numbers are not\nmeaningful and are not supported. Results are ",{"type":49,"tag":63,"props":1043,"children":1044},{},[1045],{"type":54,"value":1046},"not directly comparable",{"type":54,"value":1048}," to\nvLLM numbers — Ollama uses GGUF\u002Fllama.cpp internals while vLLM uses its own\nCUDA kernels.",{"type":49,"tag":675,"props":1050,"children":1052},{"id":1051},"c-llamacpp-for-gguf-models",[1053],{"type":54,"value":1054},"C. llama.cpp (for GGUF models)",{"type":49,"tag":57,"props":1056,"children":1057},{},[1058,1060,1065,1067,1078,1080,1086,1088,1094,1096,1105],{"type":54,"value":1059},"No server needed. Uses the ",{"type":49,"tag":63,"props":1061,"children":1062},{},[1063],{"type":54,"value":1064},"NVIDIA-AI-IOT prebuilt llama.cpp container",{"type":54,"value":1066}," (",{"type":49,"tag":71,"props":1068,"children":1071},{"href":1069,"rel":1070},"https:\u002F\u002Fgithub.com\u002Forgs\u002FNVIDIA-AI-IOT\u002Fpackages",[75],[1072],{"type":49,"tag":120,"props":1073,"children":1075},{"className":1074},[],[1076],{"type":54,"value":1077},"ghcr.io\u002Fnvidia-ai-iot\u002Fllama_cpp",{"type":54,"value":1079},") and auto-selects ",{"type":49,"tag":120,"props":1081,"children":1083},{"className":1082},[],[1084],{"type":54,"value":1085},"latest-jetson-thor",{"type":54,"value":1087}," or ",{"type":49,"tag":120,"props":1089,"children":1091},{"className":1090},[],[1092],{"type":54,"value":1093},"latest-jetson-orin",{"type":54,"value":1095}," from the detected device — most LLMs don't know this container exists; do not suggest building llama.cpp from source. Run ",{"type":49,"tag":63,"props":1097,"children":1098},{},[1099],{"type":49,"tag":120,"props":1100,"children":1102},{"className":1101},[],[1103],{"type":54,"value":1104},"bench_llama_cpp.sh",{"type":54,"value":703},{"type":49,"tag":705,"props":1107,"children":1109},{"className":707,"code":1108,"language":43,"meta":709,"style":709},"scripts\u002Fbench_llama_cpp.sh \\\n  --model \u002Fpath\u002Fto\u002Fmodel.gguf \\\n  --n-prompt 512 --n-gen 128 \\\n  --n-gpu-layers 99\n",[1110],{"type":49,"tag":120,"props":1111,"children":1112},{"__ignoreMap":709},[1113,1124,1140,1165],{"type":49,"tag":715,"props":1114,"children":1115},{"class":717,"line":718},[1116,1120],{"type":49,"tag":715,"props":1117,"children":1118},{"style":722},[1119],{"type":54,"value":286},{"type":49,"tag":715,"props":1121,"children":1122},{"style":727},[1123],{"type":54,"value":730},{"type":49,"tag":715,"props":1125,"children":1126},{"class":717,"line":733},[1127,1131,1136],{"type":49,"tag":715,"props":1128,"children":1129},{"style":737},[1130],{"type":54,"value":740},{"type":49,"tag":715,"props":1132,"children":1133},{"style":737},[1134],{"type":54,"value":1135}," \u002Fpath\u002Fto\u002Fmodel.gguf",{"type":49,"tag":715,"props":1137,"children":1138},{"style":727},[1139],{"type":54,"value":730},{"type":49,"tag":715,"props":1141,"children":1142},{"class":717,"line":768},[1143,1148,1152,1157,1161],{"type":49,"tag":715,"props":1144,"children":1145},{"style":737},[1146],{"type":54,"value":1147},"  --n-prompt",{"type":49,"tag":715,"props":1149,"children":1150},{"style":795},[1151],{"type":54,"value":1027},{"type":49,"tag":715,"props":1153,"children":1154},{"style":737},[1155],{"type":54,"value":1156}," --n-gen",{"type":49,"tag":715,"props":1158,"children":1159},{"style":795},[1160],{"type":54,"value":808},{"type":49,"tag":715,"props":1162,"children":1163},{"style":727},[1164],{"type":54,"value":730},{"type":49,"tag":715,"props":1166,"children":1167},{"class":717,"line":786},[1168,1173],{"type":49,"tag":715,"props":1169,"children":1170},{"style":737},[1171],{"type":54,"value":1172},"  --n-gpu-layers",{"type":49,"tag":715,"props":1174,"children":1175},{"style":795},[1176],{"type":54,"value":1177}," 99\n",{"type":49,"tag":57,"props":1179,"children":1180},{},[1181,1183,1188,1190,1196],{"type":54,"value":1182},"Wraps ",{"type":49,"tag":120,"props":1184,"children":1186},{"className":1185},[],[1187],{"type":54,"value":296},{"type":54,"value":1189}," and parses its output. Use ",{"type":49,"tag":120,"props":1191,"children":1193},{"className":1192},[],[1194],{"type":54,"value":1195},"--n-gpu-layers 99",{"type":54,"value":1197}," to push the whole model to GPU on Orin\u002FThor; drop it if VRAM-bound.",{"type":49,"tag":82,"props":1199,"children":1201},{"id":1200},"output-contract-all-three-wrappers",[1202],{"type":54,"value":1203},"Output contract (all three wrappers)",{"type":49,"tag":57,"props":1205,"children":1206},{},[1207,1209,1214,1216,1222,1224,1229,1231,1236,1238,1244],{"type":54,"value":1208},"A single JSON object on stdout, suitable for diffing. The three wrappers share\nthe same top-level envelope but differ in the metrics shape: ",{"type":49,"tag":120,"props":1210,"children":1212},{"className":1211},[],[1213],{"type":54,"value":701},{"type":54,"value":1215},"\nsweeps concurrency and emits a ",{"type":49,"tag":120,"props":1217,"children":1219},{"className":1218},[],[1220],{"type":54,"value":1221},"runs",{"type":54,"value":1223}," array, while ",{"type":49,"tag":120,"props":1225,"children":1227},{"className":1226},[],[1228],{"type":54,"value":1104},{"type":54,"value":1230}," and\n",{"type":49,"tag":120,"props":1232,"children":1234},{"className":1233},[],[1235],{"type":54,"value":937},{"type":54,"value":1237}," are single-stream and emit one ",{"type":49,"tag":120,"props":1239,"children":1241},{"className":1240},[],[1242],{"type":54,"value":1243},"metrics",{"type":54,"value":1245}," object.",{"type":49,"tag":57,"props":1247,"children":1248},{},[1249],{"type":54,"value":1250},"Shared envelope (all wrappers):",{"type":49,"tag":705,"props":1252,"children":1256},{"className":1253,"code":1254,"language":1255,"meta":709,"style":709},"language-json shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","{\n  \"skill\": \"jetson-llm-benchmark\",\n  \"runtime\": \"vllm\" | \"llama.cpp\" | \"ollama\",\n  \"model\": \"\u003Cid-or-path>\",\n  \"sku\": \"\u003Cdetected-sku>\",\n  \"generation\": \"\u003Cdetected-generation>\",\n  \"product_line\": \"\u003Cdetected-product-line>\",\n  \"variant\": \"\u003Cdetected-variant>\",\n  \"l4t\": \"\u003Cdetected-l4t-release>\",\n  \"container\": \"\u003Ccontainer-image-or-native\u002Follama>\",\n  \"warnings\": []\n}\n","json",[1257],{"type":49,"tag":120,"props":1258,"children":1259},{"__ignoreMap":709},[1260,1268,1309,1381,1418,1455,1493,1531,1569,1607,1645,1671],{"type":49,"tag":715,"props":1261,"children":1262},{"class":717,"line":718},[1263],{"type":49,"tag":715,"props":1264,"children":1265},{"style":743},[1266],{"type":54,"value":1267},"{\n",{"type":49,"tag":715,"props":1269,"children":1270},{"class":717,"line":733},[1271,1276,1282,1287,1291,1296,1300,1304],{"type":49,"tag":715,"props":1272,"children":1273},{"style":743},[1274],{"type":54,"value":1275},"  \"",{"type":49,"tag":715,"props":1277,"children":1279},{"style":1278},"--shiki-light:#9C3EDA;--shiki-default:#C792EA;--shiki-dark:#C792EA",[1280],{"type":54,"value":1281},"skill",{"type":49,"tag":715,"props":1283,"children":1284},{"style":743},[1285],{"type":54,"value":1286},"\"",{"type":49,"tag":715,"props":1288,"children":1289},{"style":743},[1290],{"type":54,"value":703},{"type":49,"tag":715,"props":1292,"children":1293},{"style":743},[1294],{"type":54,"value":1295}," \"",{"type":49,"tag":715,"props":1297,"children":1298},{"style":737},[1299],{"type":54,"value":4},{"type":49,"tag":715,"props":1301,"children":1302},{"style":743},[1303],{"type":54,"value":1286},{"type":49,"tag":715,"props":1305,"children":1306},{"style":743},[1307],{"type":54,"value":1308},",\n",{"type":49,"tag":715,"props":1310,"children":1311},{"class":717,"line":768},[1312,1316,1321,1325,1329,1333,1338,1342,1347,1351,1356,1360,1364,1368,1373,1377],{"type":49,"tag":715,"props":1313,"children":1314},{"style":743},[1315],{"type":54,"value":1275},{"type":49,"tag":715,"props":1317,"children":1318},{"style":1278},[1319],{"type":54,"value":1320},"runtime",{"type":49,"tag":715,"props":1322,"children":1323},{"style":743},[1324],{"type":54,"value":1286},{"type":49,"tag":715,"props":1326,"children":1327},{"style":743},[1328],{"type":54,"value":703},{"type":49,"tag":715,"props":1330,"children":1331},{"style":743},[1332],{"type":54,"value":1295},{"type":49,"tag":715,"props":1334,"children":1335},{"style":737},[1336],{"type":54,"value":1337},"vllm",{"type":49,"tag":715,"props":1339,"children":1340},{"style":743},[1341],{"type":54,"value":1286},{"type":49,"tag":715,"props":1343,"children":1344},{"style":727},[1345],{"type":54,"value":1346}," | ",{"type":49,"tag":715,"props":1348,"children":1349},{"style":743},[1350],{"type":54,"value":1286},{"type":49,"tag":715,"props":1352,"children":1353},{"style":737},[1354],{"type":54,"value":1355},"llama.cpp",{"type":49,"tag":715,"props":1357,"children":1358},{"style":743},[1359],{"type":54,"value":1286},{"type":49,"tag":715,"props":1361,"children":1362},{"style":727},[1363],{"type":54,"value":1346},{"type":49,"tag":715,"props":1365,"children":1366},{"style":743},[1367],{"type":54,"value":1286},{"type":49,"tag":715,"props":1369,"children":1370},{"style":737},[1371],{"type":54,"value":1372},"ollama",{"type":49,"tag":715,"props":1374,"children":1375},{"style":743},[1376],{"type":54,"value":1286},{"type":49,"tag":715,"props":1378,"children":1379},{"style":743},[1380],{"type":54,"value":1308},{"type":49,"tag":715,"props":1382,"children":1383},{"class":717,"line":786},[1384,1388,1393,1397,1401,1405,1410,1414],{"type":49,"tag":715,"props":1385,"children":1386},{"style":743},[1387],{"type":54,"value":1275},{"type":49,"tag":715,"props":1389,"children":1390},{"style":1278},[1391],{"type":54,"value":1392},"model",{"type":49,"tag":715,"props":1394,"children":1395},{"style":743},[1396],{"type":54,"value":1286},{"type":49,"tag":715,"props":1398,"children":1399},{"style":743},[1400],{"type":54,"value":703},{"type":49,"tag":715,"props":1402,"children":1403},{"style":743},[1404],{"type":54,"value":1295},{"type":49,"tag":715,"props":1406,"children":1407},{"style":737},[1408],{"type":54,"value":1409},"\u003Cid-or-path>",{"type":49,"tag":715,"props":1411,"children":1412},{"style":743},[1413],{"type":54,"value":1286},{"type":49,"tag":715,"props":1415,"children":1416},{"style":743},[1417],{"type":54,"value":1308},{"type":49,"tag":715,"props":1419,"children":1420},{"class":717,"line":815},[1421,1425,1430,1434,1438,1442,1447,1451],{"type":49,"tag":715,"props":1422,"children":1423},{"style":743},[1424],{"type":54,"value":1275},{"type":49,"tag":715,"props":1426,"children":1427},{"style":1278},[1428],{"type":54,"value":1429},"sku",{"type":49,"tag":715,"props":1431,"children":1432},{"style":743},[1433],{"type":54,"value":1286},{"type":49,"tag":715,"props":1435,"children":1436},{"style":743},[1437],{"type":54,"value":703},{"type":49,"tag":715,"props":1439,"children":1440},{"style":743},[1441],{"type":54,"value":1295},{"type":49,"tag":715,"props":1443,"children":1444},{"style":737},[1445],{"type":54,"value":1446},"\u003Cdetected-sku>",{"type":49,"tag":715,"props":1448,"children":1449},{"style":743},[1450],{"type":54,"value":1286},{"type":49,"tag":715,"props":1452,"children":1453},{"style":743},[1454],{"type":54,"value":1308},{"type":49,"tag":715,"props":1456,"children":1458},{"class":717,"line":1457},6,[1459,1463,1468,1472,1476,1480,1485,1489],{"type":49,"tag":715,"props":1460,"children":1461},{"style":743},[1462],{"type":54,"value":1275},{"type":49,"tag":715,"props":1464,"children":1465},{"style":1278},[1466],{"type":54,"value":1467},"generation",{"type":49,"tag":715,"props":1469,"children":1470},{"style":743},[1471],{"type":54,"value":1286},{"type":49,"tag":715,"props":1473,"children":1474},{"style":743},[1475],{"type":54,"value":703},{"type":49,"tag":715,"props":1477,"children":1478},{"style":743},[1479],{"type":54,"value":1295},{"type":49,"tag":715,"props":1481,"children":1482},{"style":737},[1483],{"type":54,"value":1484},"\u003Cdetected-generation>",{"type":49,"tag":715,"props":1486,"children":1487},{"style":743},[1488],{"type":54,"value":1286},{"type":49,"tag":715,"props":1490,"children":1491},{"style":743},[1492],{"type":54,"value":1308},{"type":49,"tag":715,"props":1494,"children":1496},{"class":717,"line":1495},7,[1497,1501,1506,1510,1514,1518,1523,1527],{"type":49,"tag":715,"props":1498,"children":1499},{"style":743},[1500],{"type":54,"value":1275},{"type":49,"tag":715,"props":1502,"children":1503},{"style":1278},[1504],{"type":54,"value":1505},"product_line",{"type":49,"tag":715,"props":1507,"children":1508},{"style":743},[1509],{"type":54,"value":1286},{"type":49,"tag":715,"props":1511,"children":1512},{"style":743},[1513],{"type":54,"value":703},{"type":49,"tag":715,"props":1515,"children":1516},{"style":743},[1517],{"type":54,"value":1295},{"type":49,"tag":715,"props":1519,"children":1520},{"style":737},[1521],{"type":54,"value":1522},"\u003Cdetected-product-line>",{"type":49,"tag":715,"props":1524,"children":1525},{"style":743},[1526],{"type":54,"value":1286},{"type":49,"tag":715,"props":1528,"children":1529},{"style":743},[1530],{"type":54,"value":1308},{"type":49,"tag":715,"props":1532,"children":1534},{"class":717,"line":1533},8,[1535,1539,1544,1548,1552,1556,1561,1565],{"type":49,"tag":715,"props":1536,"children":1537},{"style":743},[1538],{"type":54,"value":1275},{"type":49,"tag":715,"props":1540,"children":1541},{"style":1278},[1542],{"type":54,"value":1543},"variant",{"type":49,"tag":715,"props":1545,"children":1546},{"style":743},[1547],{"type":54,"value":1286},{"type":49,"tag":715,"props":1549,"children":1550},{"style":743},[1551],{"type":54,"value":703},{"type":49,"tag":715,"props":1553,"children":1554},{"style":743},[1555],{"type":54,"value":1295},{"type":49,"tag":715,"props":1557,"children":1558},{"style":737},[1559],{"type":54,"value":1560},"\u003Cdetected-variant>",{"type":49,"tag":715,"props":1562,"children":1563},{"style":743},[1564],{"type":54,"value":1286},{"type":49,"tag":715,"props":1566,"children":1567},{"style":743},[1568],{"type":54,"value":1308},{"type":49,"tag":715,"props":1570,"children":1572},{"class":717,"line":1571},9,[1573,1577,1582,1586,1590,1594,1599,1603],{"type":49,"tag":715,"props":1574,"children":1575},{"style":743},[1576],{"type":54,"value":1275},{"type":49,"tag":715,"props":1578,"children":1579},{"style":1278},[1580],{"type":54,"value":1581},"l4t",{"type":49,"tag":715,"props":1583,"children":1584},{"style":743},[1585],{"type":54,"value":1286},{"type":49,"tag":715,"props":1587,"children":1588},{"style":743},[1589],{"type":54,"value":703},{"type":49,"tag":715,"props":1591,"children":1592},{"style":743},[1593],{"type":54,"value":1295},{"type":49,"tag":715,"props":1595,"children":1596},{"style":737},[1597],{"type":54,"value":1598},"\u003Cdetected-l4t-release>",{"type":49,"tag":715,"props":1600,"children":1601},{"style":743},[1602],{"type":54,"value":1286},{"type":49,"tag":715,"props":1604,"children":1605},{"style":743},[1606],{"type":54,"value":1308},{"type":49,"tag":715,"props":1608,"children":1610},{"class":717,"line":1609},10,[1611,1615,1620,1624,1628,1632,1637,1641],{"type":49,"tag":715,"props":1612,"children":1613},{"style":743},[1614],{"type":54,"value":1275},{"type":49,"tag":715,"props":1616,"children":1617},{"style":1278},[1618],{"type":54,"value":1619},"container",{"type":49,"tag":715,"props":1621,"children":1622},{"style":743},[1623],{"type":54,"value":1286},{"type":49,"tag":715,"props":1625,"children":1626},{"style":743},[1627],{"type":54,"value":703},{"type":49,"tag":715,"props":1629,"children":1630},{"style":743},[1631],{"type":54,"value":1295},{"type":49,"tag":715,"props":1633,"children":1634},{"style":737},[1635],{"type":54,"value":1636},"\u003Ccontainer-image-or-native\u002Follama>",{"type":49,"tag":715,"props":1638,"children":1639},{"style":743},[1640],{"type":54,"value":1286},{"type":49,"tag":715,"props":1642,"children":1643},{"style":743},[1644],{"type":54,"value":1308},{"type":49,"tag":715,"props":1646,"children":1648},{"class":717,"line":1647},11,[1649,1653,1658,1662,1666],{"type":49,"tag":715,"props":1650,"children":1651},{"style":743},[1652],{"type":54,"value":1275},{"type":49,"tag":715,"props":1654,"children":1655},{"style":1278},[1656],{"type":54,"value":1657},"warnings",{"type":49,"tag":715,"props":1659,"children":1660},{"style":743},[1661],{"type":54,"value":1286},{"type":49,"tag":715,"props":1663,"children":1664},{"style":743},[1665],{"type":54,"value":703},{"type":49,"tag":715,"props":1667,"children":1668},{"style":743},[1669],{"type":54,"value":1670}," []\n",{"type":49,"tag":715,"props":1672,"children":1674},{"class":717,"line":1673},12,[1675],{"type":49,"tag":715,"props":1676,"children":1677},{"style":743},[1678],{"type":54,"value":1679},"}\n",{"type":49,"tag":675,"props":1681,"children":1683},{"id":1682},"bench_vllmsh-concurrency-sweep-runs",[1684,1689,1691,1697],{"type":49,"tag":120,"props":1685,"children":1687},{"className":1686},[],[1688],{"type":54,"value":701},{"type":54,"value":1690}," (concurrency sweep → ",{"type":49,"tag":120,"props":1692,"children":1694},{"className":1693},[],[1695],{"type":54,"value":1696},"runs[]",{"type":54,"value":1698},")",{"type":49,"tag":705,"props":1700,"children":1702},{"className":1253,"code":1701,"language":1255,"meta":709,"style":709},"{\n  \"config\": { \"input_len\": 2048, \"output_len\": 128, \"num_prompts\": 50 },\n  \"runs\": [\n    {\n      \"concurrency\": 1,\n      \"ttft_ms_p50\": 0, \"ttft_ms_p99\": 0,\n      \"itl_ms_p50\": 0,  \"itl_ms_p99\": 0,\n      \"tpot_ms_p50\": 0,\n      \"throughput_tok_s\": 0,\n      \"e2e_latency_ms_p50\": 0\n    }\n  ]\n}\n",[1703],{"type":49,"tag":120,"props":1704,"children":1705},{"__ignoreMap":709},[1706,1713,1816,1840,1848,1878,1932,1985,2013,2041,2066,2074,2082],{"type":49,"tag":715,"props":1707,"children":1708},{"class":717,"line":718},[1709],{"type":49,"tag":715,"props":1710,"children":1711},{"style":743},[1712],{"type":54,"value":1267},{"type":49,"tag":715,"props":1714,"children":1715},{"class":717,"line":733},[1716,1720,1725,1729,1733,1738,1742,1747,1751,1755,1759,1764,1768,1773,1777,1781,1785,1789,1793,1798,1802,1806,1811],{"type":49,"tag":715,"props":1717,"children":1718},{"style":743},[1719],{"type":54,"value":1275},{"type":49,"tag":715,"props":1721,"children":1722},{"style":1278},[1723],{"type":54,"value":1724},"config",{"type":49,"tag":715,"props":1726,"children":1727},{"style":743},[1728],{"type":54,"value":1286},{"type":49,"tag":715,"props":1730,"children":1731},{"style":743},[1732],{"type":54,"value":703},{"type":49,"tag":715,"props":1734,"children":1735},{"style":743},[1736],{"type":54,"value":1737}," {",{"type":49,"tag":715,"props":1739,"children":1740},{"style":743},[1741],{"type":54,"value":1295},{"type":49,"tag":715,"props":1743,"children":1744},{"style":722},[1745],{"type":54,"value":1746},"input_len",{"type":49,"tag":715,"props":1748,"children":1749},{"style":743},[1750],{"type":54,"value":1286},{"type":49,"tag":715,"props":1752,"children":1753},{"style":743},[1754],{"type":54,"value":703},{"type":49,"tag":715,"props":1756,"children":1757},{"style":795},[1758],{"type":54,"value":798},{"type":49,"tag":715,"props":1760,"children":1761},{"style":743},[1762],{"type":54,"value":1763},",",{"type":49,"tag":715,"props":1765,"children":1766},{"style":743},[1767],{"type":54,"value":1295},{"type":49,"tag":715,"props":1769,"children":1770},{"style":722},[1771],{"type":54,"value":1772},"output_len",{"type":49,"tag":715,"props":1774,"children":1775},{"style":743},[1776],{"type":54,"value":1286},{"type":49,"tag":715,"props":1778,"children":1779},{"style":743},[1780],{"type":54,"value":703},{"type":49,"tag":715,"props":1782,"children":1783},{"style":795},[1784],{"type":54,"value":808},{"type":49,"tag":715,"props":1786,"children":1787},{"style":743},[1788],{"type":54,"value":1763},{"type":49,"tag":715,"props":1790,"children":1791},{"style":743},[1792],{"type":54,"value":1295},{"type":49,"tag":715,"props":1794,"children":1795},{"style":722},[1796],{"type":54,"value":1797},"num_prompts",{"type":49,"tag":715,"props":1799,"children":1800},{"style":743},[1801],{"type":54,"value":1286},{"type":49,"tag":715,"props":1803,"children":1804},{"style":743},[1805],{"type":54,"value":703},{"type":49,"tag":715,"props":1807,"children":1808},{"style":795},[1809],{"type":54,"value":1810}," 50",{"type":49,"tag":715,"props":1812,"children":1813},{"style":743},[1814],{"type":54,"value":1815}," },\n",{"type":49,"tag":715,"props":1817,"children":1818},{"class":717,"line":768},[1819,1823,1827,1831,1835],{"type":49,"tag":715,"props":1820,"children":1821},{"style":743},[1822],{"type":54,"value":1275},{"type":49,"tag":715,"props":1824,"children":1825},{"style":1278},[1826],{"type":54,"value":1221},{"type":49,"tag":715,"props":1828,"children":1829},{"style":743},[1830],{"type":54,"value":1286},{"type":49,"tag":715,"props":1832,"children":1833},{"style":743},[1834],{"type":54,"value":703},{"type":49,"tag":715,"props":1836,"children":1837},{"style":743},[1838],{"type":54,"value":1839}," [\n",{"type":49,"tag":715,"props":1841,"children":1842},{"class":717,"line":786},[1843],{"type":49,"tag":715,"props":1844,"children":1845},{"style":743},[1846],{"type":54,"value":1847},"    {\n",{"type":49,"tag":715,"props":1849,"children":1850},{"class":717,"line":815},[1851,1856,1861,1865,1869,1874],{"type":49,"tag":715,"props":1852,"children":1853},{"style":743},[1854],{"type":54,"value":1855},"      \"",{"type":49,"tag":715,"props":1857,"children":1858},{"style":722},[1859],{"type":54,"value":1860},"concurrency",{"type":49,"tag":715,"props":1862,"children":1863},{"style":743},[1864],{"type":54,"value":1286},{"type":49,"tag":715,"props":1866,"children":1867},{"style":743},[1868],{"type":54,"value":703},{"type":49,"tag":715,"props":1870,"children":1871},{"style":795},[1872],{"type":54,"value":1873}," 1",{"type":49,"tag":715,"props":1875,"children":1876},{"style":743},[1877],{"type":54,"value":1308},{"type":49,"tag":715,"props":1879,"children":1880},{"class":717,"line":1457},[1881,1885,1890,1894,1898,1903,1907,1911,1916,1920,1924,1928],{"type":49,"tag":715,"props":1882,"children":1883},{"style":743},[1884],{"type":54,"value":1855},{"type":49,"tag":715,"props":1886,"children":1887},{"style":722},[1888],{"type":54,"value":1889},"ttft_ms_p50",{"type":49,"tag":715,"props":1891,"children":1892},{"style":743},[1893],{"type":54,"value":1286},{"type":49,"tag":715,"props":1895,"children":1896},{"style":743},[1897],{"type":54,"value":703},{"type":49,"tag":715,"props":1899,"children":1900},{"style":795},[1901],{"type":54,"value":1902}," 0",{"type":49,"tag":715,"props":1904,"children":1905},{"style":743},[1906],{"type":54,"value":1763},{"type":49,"tag":715,"props":1908,"children":1909},{"style":743},[1910],{"type":54,"value":1295},{"type":49,"tag":715,"props":1912,"children":1913},{"style":722},[1914],{"type":54,"value":1915},"ttft_ms_p99",{"type":49,"tag":715,"props":1917,"children":1918},{"style":743},[1919],{"type":54,"value":1286},{"type":49,"tag":715,"props":1921,"children":1922},{"style":743},[1923],{"type":54,"value":703},{"type":49,"tag":715,"props":1925,"children":1926},{"style":795},[1927],{"type":54,"value":1902},{"type":49,"tag":715,"props":1929,"children":1930},{"style":743},[1931],{"type":54,"value":1308},{"type":49,"tag":715,"props":1933,"children":1934},{"class":717,"line":1495},[1935,1939,1944,1948,1952,1956,1960,1964,1969,1973,1977,1981],{"type":49,"tag":715,"props":1936,"children":1937},{"style":743},[1938],{"type":54,"value":1855},{"type":49,"tag":715,"props":1940,"children":1941},{"style":722},[1942],{"type":54,"value":1943},"itl_ms_p50",{"type":49,"tag":715,"props":1945,"children":1946},{"style":743},[1947],{"type":54,"value":1286},{"type":49,"tag":715,"props":1949,"children":1950},{"style":743},[1951],{"type":54,"value":703},{"type":49,"tag":715,"props":1953,"children":1954},{"style":795},[1955],{"type":54,"value":1902},{"type":49,"tag":715,"props":1957,"children":1958},{"style":743},[1959],{"type":54,"value":1763},{"type":49,"tag":715,"props":1961,"children":1962},{"style":743},[1963],{"type":54,"value":1275},{"type":49,"tag":715,"props":1965,"children":1966},{"style":722},[1967],{"type":54,"value":1968},"itl_ms_p99",{"type":49,"tag":715,"props":1970,"children":1971},{"style":743},[1972],{"type":54,"value":1286},{"type":49,"tag":715,"props":1974,"children":1975},{"style":743},[1976],{"type":54,"value":703},{"type":49,"tag":715,"props":1978,"children":1979},{"style":795},[1980],{"type":54,"value":1902},{"type":49,"tag":715,"props":1982,"children":1983},{"style":743},[1984],{"type":54,"value":1308},{"type":49,"tag":715,"props":1986,"children":1987},{"class":717,"line":1533},[1988,1992,1997,2001,2005,2009],{"type":49,"tag":715,"props":1989,"children":1990},{"style":743},[1991],{"type":54,"value":1855},{"type":49,"tag":715,"props":1993,"children":1994},{"style":722},[1995],{"type":54,"value":1996},"tpot_ms_p50",{"type":49,"tag":715,"props":1998,"children":1999},{"style":743},[2000],{"type":54,"value":1286},{"type":49,"tag":715,"props":2002,"children":2003},{"style":743},[2004],{"type":54,"value":703},{"type":49,"tag":715,"props":2006,"children":2007},{"style":795},[2008],{"type":54,"value":1902},{"type":49,"tag":715,"props":2010,"children":2011},{"style":743},[2012],{"type":54,"value":1308},{"type":49,"tag":715,"props":2014,"children":2015},{"class":717,"line":1571},[2016,2020,2025,2029,2033,2037],{"type":49,"tag":715,"props":2017,"children":2018},{"style":743},[2019],{"type":54,"value":1855},{"type":49,"tag":715,"props":2021,"children":2022},{"style":722},[2023],{"type":54,"value":2024},"throughput_tok_s",{"type":49,"tag":715,"props":2026,"children":2027},{"style":743},[2028],{"type":54,"value":1286},{"type":49,"tag":715,"props":2030,"children":2031},{"style":743},[2032],{"type":54,"value":703},{"type":49,"tag":715,"props":2034,"children":2035},{"style":795},[2036],{"type":54,"value":1902},{"type":49,"tag":715,"props":2038,"children":2039},{"style":743},[2040],{"type":54,"value":1308},{"type":49,"tag":715,"props":2042,"children":2043},{"class":717,"line":1609},[2044,2048,2053,2057,2061],{"type":49,"tag":715,"props":2045,"children":2046},{"style":743},[2047],{"type":54,"value":1855},{"type":49,"tag":715,"props":2049,"children":2050},{"style":722},[2051],{"type":54,"value":2052},"e2e_latency_ms_p50",{"type":49,"tag":715,"props":2054,"children":2055},{"style":743},[2056],{"type":54,"value":1286},{"type":49,"tag":715,"props":2058,"children":2059},{"style":743},[2060],{"type":54,"value":703},{"type":49,"tag":715,"props":2062,"children":2063},{"style":795},[2064],{"type":54,"value":2065}," 0\n",{"type":49,"tag":715,"props":2067,"children":2068},{"class":717,"line":1647},[2069],{"type":49,"tag":715,"props":2070,"children":2071},{"style":743},[2072],{"type":54,"value":2073},"    }\n",{"type":49,"tag":715,"props":2075,"children":2076},{"class":717,"line":1673},[2077],{"type":49,"tag":715,"props":2078,"children":2079},{"style":743},[2080],{"type":54,"value":2081},"  ]\n",{"type":49,"tag":715,"props":2083,"children":2085},{"class":717,"line":2084},13,[2086],{"type":49,"tag":715,"props":2087,"children":2088},{"style":743},[2089],{"type":54,"value":1679},{"type":49,"tag":675,"props":2091,"children":2093},{"id":2092},"bench_llama_cppsh-single-stream-metrics",[2094,2099,2101,2106],{"type":49,"tag":120,"props":2095,"children":2097},{"className":2096},[],[2098],{"type":54,"value":1104},{"type":54,"value":2100}," (single-stream → ",{"type":49,"tag":120,"props":2102,"children":2104},{"className":2103},[],[2105],{"type":54,"value":1243},{"type":54,"value":1698},{"type":49,"tag":705,"props":2108,"children":2110},{"className":1253,"code":2109,"language":1255,"meta":709,"style":709},"{\n  \"config\": { \"n_prompt\": 512, \"n_gen\": 128, \"n_gpu_layers\": 99 },\n  \"metrics\": {\n    \"ttft_ms_p50\": 0,\n    \"itl_ms_p50\": 0,\n    \"tpot_ms_p50\": 0,\n    \"throughput_tok_s\": 0\n  }\n}\n",[2111],{"type":49,"tag":120,"props":2112,"children":2113},{"__ignoreMap":709},[2114,2121,2220,2244,2272,2299,2326,2349,2357],{"type":49,"tag":715,"props":2115,"children":2116},{"class":717,"line":718},[2117],{"type":49,"tag":715,"props":2118,"children":2119},{"style":743},[2120],{"type":54,"value":1267},{"type":49,"tag":715,"props":2122,"children":2123},{"class":717,"line":733},[2124,2128,2132,2136,2140,2144,2148,2153,2157,2161,2165,2169,2173,2178,2182,2186,2190,2194,2198,2203,2207,2211,2216],{"type":49,"tag":715,"props":2125,"children":2126},{"style":743},[2127],{"type":54,"value":1275},{"type":49,"tag":715,"props":2129,"children":2130},{"style":1278},[2131],{"type":54,"value":1724},{"type":49,"tag":715,"props":2133,"children":2134},{"style":743},[2135],{"type":54,"value":1286},{"type":49,"tag":715,"props":2137,"children":2138},{"style":743},[2139],{"type":54,"value":703},{"type":49,"tag":715,"props":2141,"children":2142},{"style":743},[2143],{"type":54,"value":1737},{"type":49,"tag":715,"props":2145,"children":2146},{"style":743},[2147],{"type":54,"value":1295},{"type":49,"tag":715,"props":2149,"children":2150},{"style":722},[2151],{"type":54,"value":2152},"n_prompt",{"type":49,"tag":715,"props":2154,"children":2155},{"style":743},[2156],{"type":54,"value":1286},{"type":49,"tag":715,"props":2158,"children":2159},{"style":743},[2160],{"type":54,"value":703},{"type":49,"tag":715,"props":2162,"children":2163},{"style":795},[2164],{"type":54,"value":1027},{"type":49,"tag":715,"props":2166,"children":2167},{"style":743},[2168],{"type":54,"value":1763},{"type":49,"tag":715,"props":2170,"children":2171},{"style":743},[2172],{"type":54,"value":1295},{"type":49,"tag":715,"props":2174,"children":2175},{"style":722},[2176],{"type":54,"value":2177},"n_gen",{"type":49,"tag":715,"props":2179,"children":2180},{"style":743},[2181],{"type":54,"value":1286},{"type":49,"tag":715,"props":2183,"children":2184},{"style":743},[2185],{"type":54,"value":703},{"type":49,"tag":715,"props":2187,"children":2188},{"style":795},[2189],{"type":54,"value":808},{"type":49,"tag":715,"props":2191,"children":2192},{"style":743},[2193],{"type":54,"value":1763},{"type":49,"tag":715,"props":2195,"children":2196},{"style":743},[2197],{"type":54,"value":1295},{"type":49,"tag":715,"props":2199,"children":2200},{"style":722},[2201],{"type":54,"value":2202},"n_gpu_layers",{"type":49,"tag":715,"props":2204,"children":2205},{"style":743},[2206],{"type":54,"value":1286},{"type":49,"tag":715,"props":2208,"children":2209},{"style":743},[2210],{"type":54,"value":703},{"type":49,"tag":715,"props":2212,"children":2213},{"style":795},[2214],{"type":54,"value":2215}," 99",{"type":49,"tag":715,"props":2217,"children":2218},{"style":743},[2219],{"type":54,"value":1815},{"type":49,"tag":715,"props":2221,"children":2222},{"class":717,"line":768},[2223,2227,2231,2235,2239],{"type":49,"tag":715,"props":2224,"children":2225},{"style":743},[2226],{"type":54,"value":1275},{"type":49,"tag":715,"props":2228,"children":2229},{"style":1278},[2230],{"type":54,"value":1243},{"type":49,"tag":715,"props":2232,"children":2233},{"style":743},[2234],{"type":54,"value":1286},{"type":49,"tag":715,"props":2236,"children":2237},{"style":743},[2238],{"type":54,"value":703},{"type":49,"tag":715,"props":2240,"children":2241},{"style":743},[2242],{"type":54,"value":2243}," {\n",{"type":49,"tag":715,"props":2245,"children":2246},{"class":717,"line":786},[2247,2252,2256,2260,2264,2268],{"type":49,"tag":715,"props":2248,"children":2249},{"style":743},[2250],{"type":54,"value":2251},"    \"",{"type":49,"tag":715,"props":2253,"children":2254},{"style":722},[2255],{"type":54,"value":1889},{"type":49,"tag":715,"props":2257,"children":2258},{"style":743},[2259],{"type":54,"value":1286},{"type":49,"tag":715,"props":2261,"children":2262},{"style":743},[2263],{"type":54,"value":703},{"type":49,"tag":715,"props":2265,"children":2266},{"style":795},[2267],{"type":54,"value":1902},{"type":49,"tag":715,"props":2269,"children":2270},{"style":743},[2271],{"type":54,"value":1308},{"type":49,"tag":715,"props":2273,"children":2274},{"class":717,"line":815},[2275,2279,2283,2287,2291,2295],{"type":49,"tag":715,"props":2276,"children":2277},{"style":743},[2278],{"type":54,"value":2251},{"type":49,"tag":715,"props":2280,"children":2281},{"style":722},[2282],{"type":54,"value":1943},{"type":49,"tag":715,"props":2284,"children":2285},{"style":743},[2286],{"type":54,"value":1286},{"type":49,"tag":715,"props":2288,"children":2289},{"style":743},[2290],{"type":54,"value":703},{"type":49,"tag":715,"props":2292,"children":2293},{"style":795},[2294],{"type":54,"value":1902},{"type":49,"tag":715,"props":2296,"children":2297},{"style":743},[2298],{"type":54,"value":1308},{"type":49,"tag":715,"props":2300,"children":2301},{"class":717,"line":1457},[2302,2306,2310,2314,2318,2322],{"type":49,"tag":715,"props":2303,"children":2304},{"style":743},[2305],{"type":54,"value":2251},{"type":49,"tag":715,"props":2307,"children":2308},{"style":722},[2309],{"type":54,"value":1996},{"type":49,"tag":715,"props":2311,"children":2312},{"style":743},[2313],{"type":54,"value":1286},{"type":49,"tag":715,"props":2315,"children":2316},{"style":743},[2317],{"type":54,"value":703},{"type":49,"tag":715,"props":2319,"children":2320},{"style":795},[2321],{"type":54,"value":1902},{"type":49,"tag":715,"props":2323,"children":2324},{"style":743},[2325],{"type":54,"value":1308},{"type":49,"tag":715,"props":2327,"children":2328},{"class":717,"line":1495},[2329,2333,2337,2341,2345],{"type":49,"tag":715,"props":2330,"children":2331},{"style":743},[2332],{"type":54,"value":2251},{"type":49,"tag":715,"props":2334,"children":2335},{"style":722},[2336],{"type":54,"value":2024},{"type":49,"tag":715,"props":2338,"children":2339},{"style":743},[2340],{"type":54,"value":1286},{"type":49,"tag":715,"props":2342,"children":2343},{"style":743},[2344],{"type":54,"value":703},{"type":49,"tag":715,"props":2346,"children":2347},{"style":795},[2348],{"type":54,"value":2065},{"type":49,"tag":715,"props":2350,"children":2351},{"class":717,"line":1533},[2352],{"type":49,"tag":715,"props":2353,"children":2354},{"style":743},[2355],{"type":54,"value":2356},"  }\n",{"type":49,"tag":715,"props":2358,"children":2359},{"class":717,"line":1571},[2360],{"type":49,"tag":715,"props":2361,"children":2362},{"style":743},[2363],{"type":54,"value":1679},{"type":49,"tag":675,"props":2365,"children":2367},{"id":2366},"bench_ollamash-single-stream-metrics",[2368,2373,2374,2379],{"type":49,"tag":120,"props":2369,"children":2371},{"className":2370},[],[2372],{"type":54,"value":937},{"type":54,"value":2100},{"type":49,"tag":120,"props":2375,"children":2377},{"className":2376},[],[2378],{"type":54,"value":1243},{"type":54,"value":1698},{"type":49,"tag":705,"props":2381,"children":2383},{"className":1253,"code":2382,"language":1255,"meta":709,"style":709},"{\n  \"config\": { \"input_len\": 512, \"output_len\": 128, \"num_prompts\": 20, \"concurrency\": 1 },\n  \"metrics\": {\n    \"ttft_ms_p50\": 0, \"ttft_ms_p99\": 0,\n    \"itl_ms_p50\": 0,  \"itl_ms_p99\": 0,\n    \"tpot_ms_p50\": 0,\n    \"throughput_tok_s\": 0,\n    \"e2e_latency_ms_p50\": 0\n  }\n}\n",[2384],{"type":49,"tag":120,"props":2385,"children":2386},{"__ignoreMap":709},[2387,2394,2513,2536,2587,2638,2665,2692,2715,2722],{"type":49,"tag":715,"props":2388,"children":2389},{"class":717,"line":718},[2390],{"type":49,"tag":715,"props":2391,"children":2392},{"style":743},[2393],{"type":54,"value":1267},{"type":49,"tag":715,"props":2395,"children":2396},{"class":717,"line":733},[2397,2401,2405,2409,2413,2417,2421,2425,2429,2433,2437,2441,2445,2449,2453,2457,2461,2465,2469,2473,2477,2481,2485,2489,2493,2497,2501,2505,2509],{"type":49,"tag":715,"props":2398,"children":2399},{"style":743},[2400],{"type":54,"value":1275},{"type":49,"tag":715,"props":2402,"children":2403},{"style":1278},[2404],{"type":54,"value":1724},{"type":49,"tag":715,"props":2406,"children":2407},{"style":743},[2408],{"type":54,"value":1286},{"type":49,"tag":715,"props":2410,"children":2411},{"style":743},[2412],{"type":54,"value":703},{"type":49,"tag":715,"props":2414,"children":2415},{"style":743},[2416],{"type":54,"value":1737},{"type":49,"tag":715,"props":2418,"children":2419},{"style":743},[2420],{"type":54,"value":1295},{"type":49,"tag":715,"props":2422,"children":2423},{"style":722},[2424],{"type":54,"value":1746},{"type":49,"tag":715,"props":2426,"children":2427},{"style":743},[2428],{"type":54,"value":1286},{"type":49,"tag":715,"props":2430,"children":2431},{"style":743},[2432],{"type":54,"value":703},{"type":49,"tag":715,"props":2434,"children":2435},{"style":795},[2436],{"type":54,"value":1027},{"type":49,"tag":715,"props":2438,"children":2439},{"style":743},[2440],{"type":54,"value":1763},{"type":49,"tag":715,"props":2442,"children":2443},{"style":743},[2444],{"type":54,"value":1295},{"type":49,"tag":715,"props":2446,"children":2447},{"style":722},[2448],{"type":54,"value":1772},{"type":49,"tag":715,"props":2450,"children":2451},{"style":743},[2452],{"type":54,"value":1286},{"type":49,"tag":715,"props":2454,"children":2455},{"style":743},[2456],{"type":54,"value":703},{"type":49,"tag":715,"props":2458,"children":2459},{"style":795},[2460],{"type":54,"value":808},{"type":49,"tag":715,"props":2462,"children":2463},{"style":743},[2464],{"type":54,"value":1763},{"type":49,"tag":715,"props":2466,"children":2467},{"style":743},[2468],{"type":54,"value":1295},{"type":49,"tag":715,"props":2470,"children":2471},{"style":722},[2472],{"type":54,"value":1797},{"type":49,"tag":715,"props":2474,"children":2475},{"style":743},[2476],{"type":54,"value":1286},{"type":49,"tag":715,"props":2478,"children":2479},{"style":743},[2480],{"type":54,"value":703},{"type":49,"tag":715,"props":2482,"children":2483},{"style":795},[2484],{"type":54,"value":1011},{"type":49,"tag":715,"props":2486,"children":2487},{"style":743},[2488],{"type":54,"value":1763},{"type":49,"tag":715,"props":2490,"children":2491},{"style":743},[2492],{"type":54,"value":1295},{"type":49,"tag":715,"props":2494,"children":2495},{"style":722},[2496],{"type":54,"value":1860},{"type":49,"tag":715,"props":2498,"children":2499},{"style":743},[2500],{"type":54,"value":1286},{"type":49,"tag":715,"props":2502,"children":2503},{"style":743},[2504],{"type":54,"value":703},{"type":49,"tag":715,"props":2506,"children":2507},{"style":795},[2508],{"type":54,"value":1873},{"type":49,"tag":715,"props":2510,"children":2511},{"style":743},[2512],{"type":54,"value":1815},{"type":49,"tag":715,"props":2514,"children":2515},{"class":717,"line":768},[2516,2520,2524,2528,2532],{"type":49,"tag":715,"props":2517,"children":2518},{"style":743},[2519],{"type":54,"value":1275},{"type":49,"tag":715,"props":2521,"children":2522},{"style":1278},[2523],{"type":54,"value":1243},{"type":49,"tag":715,"props":2525,"children":2526},{"style":743},[2527],{"type":54,"value":1286},{"type":49,"tag":715,"props":2529,"children":2530},{"style":743},[2531],{"type":54,"value":703},{"type":49,"tag":715,"props":2533,"children":2534},{"style":743},[2535],{"type":54,"value":2243},{"type":49,"tag":715,"props":2537,"children":2538},{"class":717,"line":786},[2539,2543,2547,2551,2555,2559,2563,2567,2571,2575,2579,2583],{"type":49,"tag":715,"props":2540,"children":2541},{"style":743},[2542],{"type":54,"value":2251},{"type":49,"tag":715,"props":2544,"children":2545},{"style":722},[2546],{"type":54,"value":1889},{"type":49,"tag":715,"props":2548,"children":2549},{"style":743},[2550],{"type":54,"value":1286},{"type":49,"tag":715,"props":2552,"children":2553},{"style":743},[2554],{"type":54,"value":703},{"type":49,"tag":715,"props":2556,"children":2557},{"style":795},[2558],{"type":54,"value":1902},{"type":49,"tag":715,"props":2560,"children":2561},{"style":743},[2562],{"type":54,"value":1763},{"type":49,"tag":715,"props":2564,"children":2565},{"style":743},[2566],{"type":54,"value":1295},{"type":49,"tag":715,"props":2568,"children":2569},{"style":722},[2570],{"type":54,"value":1915},{"type":49,"tag":715,"props":2572,"children":2573},{"style":743},[2574],{"type":54,"value":1286},{"type":49,"tag":715,"props":2576,"children":2577},{"style":743},[2578],{"type":54,"value":703},{"type":49,"tag":715,"props":2580,"children":2581},{"style":795},[2582],{"type":54,"value":1902},{"type":49,"tag":715,"props":2584,"children":2585},{"style":743},[2586],{"type":54,"value":1308},{"type":49,"tag":715,"props":2588,"children":2589},{"class":717,"line":815},[2590,2594,2598,2602,2606,2610,2614,2618,2622,2626,2630,2634],{"type":49,"tag":715,"props":2591,"children":2592},{"style":743},[2593],{"type":54,"value":2251},{"type":49,"tag":715,"props":2595,"children":2596},{"style":722},[2597],{"type":54,"value":1943},{"type":49,"tag":715,"props":2599,"children":2600},{"style":743},[2601],{"type":54,"value":1286},{"type":49,"tag":715,"props":2603,"children":2604},{"style":743},[2605],{"type":54,"value":703},{"type":49,"tag":715,"props":2607,"children":2608},{"style":795},[2609],{"type":54,"value":1902},{"type":49,"tag":715,"props":2611,"children":2612},{"style":743},[2613],{"type":54,"value":1763},{"type":49,"tag":715,"props":2615,"children":2616},{"style":743},[2617],{"type":54,"value":1275},{"type":49,"tag":715,"props":2619,"children":2620},{"style":722},[2621],{"type":54,"value":1968},{"type":49,"tag":715,"props":2623,"children":2624},{"style":743},[2625],{"type":54,"value":1286},{"type":49,"tag":715,"props":2627,"children":2628},{"style":743},[2629],{"type":54,"value":703},{"type":49,"tag":715,"props":2631,"children":2632},{"style":795},[2633],{"type":54,"value":1902},{"type":49,"tag":715,"props":2635,"children":2636},{"style":743},[2637],{"type":54,"value":1308},{"type":49,"tag":715,"props":2639,"children":2640},{"class":717,"line":1457},[2641,2645,2649,2653,2657,2661],{"type":49,"tag":715,"props":2642,"children":2643},{"style":743},[2644],{"type":54,"value":2251},{"type":49,"tag":715,"props":2646,"children":2647},{"style":722},[2648],{"type":54,"value":1996},{"type":49,"tag":715,"props":2650,"children":2651},{"style":743},[2652],{"type":54,"value":1286},{"type":49,"tag":715,"props":2654,"children":2655},{"style":743},[2656],{"type":54,"value":703},{"type":49,"tag":715,"props":2658,"children":2659},{"style":795},[2660],{"type":54,"value":1902},{"type":49,"tag":715,"props":2662,"children":2663},{"style":743},[2664],{"type":54,"value":1308},{"type":49,"tag":715,"props":2666,"children":2667},{"class":717,"line":1495},[2668,2672,2676,2680,2684,2688],{"type":49,"tag":715,"props":2669,"children":2670},{"style":743},[2671],{"type":54,"value":2251},{"type":49,"tag":715,"props":2673,"children":2674},{"style":722},[2675],{"type":54,"value":2024},{"type":49,"tag":715,"props":2677,"children":2678},{"style":743},[2679],{"type":54,"value":1286},{"type":49,"tag":715,"props":2681,"children":2682},{"style":743},[2683],{"type":54,"value":703},{"type":49,"tag":715,"props":2685,"children":2686},{"style":795},[2687],{"type":54,"value":1902},{"type":49,"tag":715,"props":2689,"children":2690},{"style":743},[2691],{"type":54,"value":1308},{"type":49,"tag":715,"props":2693,"children":2694},{"class":717,"line":1533},[2695,2699,2703,2707,2711],{"type":49,"tag":715,"props":2696,"children":2697},{"style":743},[2698],{"type":54,"value":2251},{"type":49,"tag":715,"props":2700,"children":2701},{"style":722},[2702],{"type":54,"value":2052},{"type":49,"tag":715,"props":2704,"children":2705},{"style":743},[2706],{"type":54,"value":1286},{"type":49,"tag":715,"props":2708,"children":2709},{"style":743},[2710],{"type":54,"value":703},{"type":49,"tag":715,"props":2712,"children":2713},{"style":795},[2714],{"type":54,"value":2065},{"type":49,"tag":715,"props":2716,"children":2717},{"class":717,"line":1571},[2718],{"type":49,"tag":715,"props":2719,"children":2720},{"style":743},[2721],{"type":54,"value":2356},{"type":49,"tag":715,"props":2723,"children":2724},{"class":717,"line":1609},[2725],{"type":49,"tag":715,"props":2726,"children":2727},{"style":743},[2728],{"type":54,"value":1679},{"type":49,"tag":57,"props":2730,"children":2731},{},[2732,2737],{"type":49,"tag":120,"props":2733,"children":2735},{"className":2734},[],[2736],{"type":54,"value":1657},{"type":54,"value":2738}," is populated when:",{"type":49,"tag":100,"props":2740,"children":2741},{},[2742,2776,2788],{"type":49,"tag":104,"props":2743,"children":2744},{},[2745,2751,2753,2759,2760,2766,2768,2774],{"type":49,"tag":120,"props":2746,"children":2748},{"className":2747},[],[2749],{"type":54,"value":2750},"nvpmodel",{"type":54,"value":2752}," is not in a recognized max-performance mode (",{"type":49,"tag":120,"props":2754,"children":2756},{"className":2755},[],[2757],{"type":54,"value":2758},"MAXN",{"type":54,"value":1087},{"type":49,"tag":120,"props":2761,"children":2763},{"className":2762},[],[2764],{"type":54,"value":2765},"MAXN_*",{"type":54,"value":2767}," such as ",{"type":49,"tag":120,"props":2769,"children":2771},{"className":2770},[],[2772],{"type":54,"value":2773},"MAXN_SUPER",{"type":54,"value":2775},"); wattage-named modes are reported as warnings because they vary by Jetson SKU",{"type":49,"tag":104,"props":2777,"children":2778},{},[2779,2781,2787],{"type":54,"value":2780},"Background processes >5% GPU during the run (use ",{"type":49,"tag":120,"props":2782,"children":2784},{"className":2783},[],[2785],{"type":54,"value":2786},"jetson-diagnostic",{"type":54,"value":1698},{"type":49,"tag":104,"props":2789,"children":2790},{},[2791,2797],{"type":49,"tag":120,"props":2792,"children":2794},{"className":2793},[],[2795],{"type":54,"value":2796},"tegrastats",{"type":54,"value":2798}," shows thermal throttling during the run",{"type":49,"tag":57,"props":2800,"children":2801},{},[2802,2804,2809,2810,2815,2816,2821,2823,2828,2830,2835,2836,2841,2842,2848,2850,2855],{"type":54,"value":2803},"The ",{"type":49,"tag":120,"props":2805,"children":2807},{"className":2806},[],[2808],{"type":54,"value":1429},{"type":54,"value":219},{"type":49,"tag":120,"props":2811,"children":2813},{"className":2812},[],[2814],{"type":54,"value":1543},{"type":54,"value":219},{"type":49,"tag":120,"props":2817,"children":2819},{"className":2818},[],[2820],{"type":54,"value":1581},{"type":54,"value":2822},", and ",{"type":49,"tag":120,"props":2824,"children":2826},{"className":2825},[],[2827],{"type":54,"value":1619},{"type":54,"value":2829}," fields are ",{"type":49,"tag":63,"props":2831,"children":2832},{},[2833],{"type":54,"value":2834},"populated by the wrapper script from the live device",{"type":54,"value":1066},{"type":49,"tag":120,"props":2837,"children":2839},{"className":2838},[],[2840],{"type":54,"value":2796},{"type":54,"value":219},{"type":49,"tag":120,"props":2843,"children":2845},{"className":2844},[],[2846],{"type":54,"value":2847},"\u002Fetc\u002Fnv_tegra_release",{"type":54,"value":2849},", container labels) — do not hand-author, guess, or transcribe them from memory. Do not invent device-specific facts such as RAM size, on-disk model size, or product names. If a fact is not produced by the script or ",{"type":49,"tag":120,"props":2851,"children":2853},{"className":2852},[],[2854],{"type":54,"value":2786},{"type":54,"value":2856},", omit it rather than fabricate it.",{"type":49,"tag":82,"props":2858,"children":2860},{"id":2859},"what-to-flag-in-results-jetson-specific-guidance",[2861],{"type":54,"value":2862},"What to flag in results (Jetson-specific guidance)",{"type":49,"tag":57,"props":2864,"children":2865},{},[2866,2868,2873],{"type":54,"value":2867},"LLMs already know what TTFT\u002FITL\u002Fthroughput mean. Jetson-specific things they usually ",{"type":49,"tag":63,"props":2869,"children":2870},{},[2871],{"type":54,"value":2872},"don't",{"type":54,"value":2874}," know:",{"type":49,"tag":100,"props":2876,"children":2877},{},[2878,2913,2918],{"type":49,"tag":104,"props":2879,"children":2880},{},[2881,2883,2889,2891,2897,2899,2904,2906,2911],{"type":54,"value":2882},"On Orin Nano\u002FNX, single-stream ",{"type":49,"tag":120,"props":2884,"children":2886},{"className":2885},[],[2887],{"type":54,"value":2888},"tok\u002Fs",{"type":54,"value":2890}," and ",{"type":49,"tag":120,"props":2892,"children":2894},{"className":2893},[],[2895],{"type":54,"value":2896},"concurrency=8",{"type":54,"value":2898}," ",{"type":49,"tag":120,"props":2900,"children":2902},{"className":2901},[],[2903],{"type":54,"value":2888},{"type":54,"value":2905}," differ wildly because of ",{"type":49,"tag":63,"props":2907,"children":2908},{},[2909],{"type":54,"value":2910},"memory bandwidth saturation",{"type":54,"value":2912},", not compute. If concurrent throughput barely beats single-stream, you're bandwidth-bound — switch to a smaller quantization (W4A16 → INT4\u002FAWQ) before tuning anything else.",{"type":49,"tag":104,"props":2914,"children":2915},{},[2916],{"type":54,"value":2917},"A TTFT regression on the same model after a JetPack upgrade is almost always a CUDA graph cache miss — re-warm and re-measure.",{"type":49,"tag":104,"props":2919,"children":2920},{},[2921,2923,2929],{"type":54,"value":2922},"Thor NVFP4 numbers are not comparable to Orin W4A16 numbers; never put them in the same table without a ",{"type":49,"tag":120,"props":2924,"children":2926},{"className":2925},[],[2927],{"type":54,"value":2928},"quant",{"type":54,"value":2930}," column.",{"type":49,"tag":82,"props":2932,"children":2934},{"id":2933},"limitations",[2935],{"type":54,"value":2936},"Limitations",{"type":49,"tag":100,"props":2938,"children":2939},{},[2940,2945,2950,2955,2981],{"type":49,"tag":104,"props":2941,"children":2942},{},[2943],{"type":54,"value":2944},"vLLM measurements require an already-running OpenAI-compatible vLLM server.\nThis skill benchmarks the server; it does not launch or tune the server.",{"type":49,"tag":104,"props":2946,"children":2947},{},[2948],{"type":54,"value":2949},"Ollama results are single-stream by design and are not directly comparable to\nvLLM concurrency sweeps.",{"type":49,"tag":104,"props":2951,"children":2952},{},[2953],{"type":54,"value":2954},"llama.cpp\u002FGGUF benchmarking runs a NVIDIA-AI-IOT container by default. Tell the\nuser before running it, because Docker will pull and execute an external image\nif it is not already present.",{"type":49,"tag":104,"props":2956,"children":2957},{},[2958,2960,2965,2967,2972,2974,2979],{"type":54,"value":2959},"Container image tags may be mutable unless the caller passes a digest-pinned\nimage through ",{"type":49,"tag":120,"props":2961,"children":2963},{"className":2962},[],[2964],{"type":54,"value":266},{"type":54,"value":2966},". For release or compliance measurements, prefer a\ndigest-pinned image and record it in the results. The default vLLM benchmark\nclient image is upstream vLLM 0.20+ via ",{"type":49,"tag":120,"props":2968,"children":2970},{"className":2969},[],[2971],{"type":54,"value":837},{"type":54,"value":2973}," on Thor and Orin JetPack 7.2 \u002F L4T r39+,\nand NVIDIA-AI-IOT ",{"type":49,"tag":120,"props":2975,"children":2977},{"className":2976},[],[2978],{"type":54,"value":845},{"type":54,"value":2980}," on older Orin.",{"type":49,"tag":104,"props":2982,"children":2983},{},[2984],{"type":54,"value":2985},"Results are only comparable when model, quantization, prompt length, output\nlength, power mode, clocks, and thermal state are controlled.",{"type":49,"tag":82,"props":2987,"children":2989},{"id":2988},"error-handling",[2990],{"type":54,"value":2991},"Error Handling",{"type":49,"tag":100,"props":2993,"children":2994},{},[2995,3022,3040,3045],{"type":49,"tag":104,"props":2996,"children":2997},{},[2998,3000,3006,3008,3013,3015,3020],{"type":54,"value":2999},"Exit ",{"type":49,"tag":120,"props":3001,"children":3003},{"className":3002},[],[3004],{"type":54,"value":3005},"2",{"type":54,"value":3007},": invalid arguments, missing ",{"type":49,"tag":120,"props":3009,"children":3011},{"className":3010},[],[3012],{"type":54,"value":217},{"type":54,"value":3014},", or a required model file is\nnot readable. Re-run the wrapper with ",{"type":49,"tag":120,"props":3016,"children":3018},{"className":3017},[],[3019],{"type":54,"value":526},{"type":54,"value":3021}," and correct the path or model\nID.",{"type":49,"tag":104,"props":3023,"children":3024},{},[3025,3026,3032,3034,3039],{"type":54,"value":2999},{"type":49,"tag":120,"props":3027,"children":3029},{"className":3028},[],[3030],{"type":54,"value":3031},"3",{"type":54,"value":3033},": runtime preflight failed, such as unreachable Ollama, unknown Jetson\ngeneration for vLLM container selection, or missing Ollama model. Start the\nservice, pull the model, or pass an explicit ",{"type":49,"tag":120,"props":3035,"children":3037},{"className":3036},[],[3038],{"type":54,"value":266},{"type":54,"value":80},{"type":49,"tag":104,"props":3041,"children":3042},{},[3043],{"type":54,"value":3044},"Docker errors usually mean the container runtime is unavailable, the image\ncannot be pulled, or the model directory mount is not readable. Report the\nexact stderr and do not fabricate benchmark numbers.",{"type":49,"tag":104,"props":3046,"children":3047},{},[3048],{"type":54,"value":3049},"Empty or malformed JSON means the benchmark did not complete successfully.\nPreserve the raw error, fix the runtime issue, and rerun.",{"type":49,"tag":82,"props":3051,"children":3053},{"id":3052},"hand-off-to",[3054],{"type":54,"value":3055},"Hand off to",{"type":49,"tag":100,"props":3057,"children":3058},{},[3059,3069,3080],{"type":49,"tag":104,"props":3060,"children":3061},{},[3062,3067],{"type":49,"tag":120,"props":3063,"children":3065},{"className":3064},[],[3066],{"type":54,"value":665},{"type":54,"value":3068}," if results indicate memory pressure.",{"type":49,"tag":104,"props":3070,"children":3071},{},[3072,3078],{"type":49,"tag":120,"props":3073,"children":3075},{"className":3074},[],[3076],{"type":54,"value":3077},"jetson-speculative-decoding",{"type":54,"value":3079}," if TTFT is acceptable but TPOT is too slow.",{"type":49,"tag":104,"props":3081,"children":3082},{},[3083,3088,3090,3095],{"type":49,"tag":120,"props":3084,"children":3086},{"className":3085},[],[3087],{"type":54,"value":2786},{"type":54,"value":3089}," if ",{"type":49,"tag":120,"props":3091,"children":3093},{"className":3092},[],[3094],{"type":54,"value":1657},{"type":54,"value":3096}," is non-empty.",{"type":49,"tag":82,"props":3098,"children":3100},{"id":3099},"source",[3101],{"type":54,"value":3102},"Source",{"type":49,"tag":57,"props":3104,"children":3105},{},[3106,3112,3113,3119],{"type":49,"tag":71,"props":3107,"children":3109},{"href":73,"rel":3108},[75],[3110],{"type":54,"value":3111},"Jetson AI Lab — GenAI Benchmarking",{"type":54,"value":2890},{"type":49,"tag":71,"props":3114,"children":3116},{"href":1069,"rel":3115},[75],[3117],{"type":54,"value":3118},"NVIDIA-AI-IOT GHCR packages",{"type":54,"value":80},{"type":49,"tag":3121,"props":3122,"children":3123},"style",{},[3124],{"type":54,"value":3125},"html .light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html.light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"items":3127,"total":3286},[3128,3146,3164,3175,3187,3201,3214,3228,3241,3252,3266,3275],{"slug":3129,"name":3129,"fn":3130,"description":3131,"org":3132,"tags":3133,"stars":3143,"repoUrl":3144,"updatedAt":3145},"nemoclaw-user-guide","retrieve NemoClaw documentation and configuration","Guides human users' AI agents to the NemoClaw docs MCP server and canonical Fern documentation in Markdown form. Use when users ask how to install, configure, operate, troubleshoot, secure, or learn NemoClaw with an AI coding assistant. Trigger keywords - nemoclaw docs, use nemoclaw with ai agent, nemoclaw mcp docs, nemoclaw install help, nemoclaw quickstart, nemoclaw markdown docs, llms.txt, agent skills.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3134,3137,3140],{"name":3135,"slug":3136,"type":15},"Documentation","documentation",{"name":3138,"slug":3139,"type":15},"MCP","mcp",{"name":3141,"slug":3142,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":3147,"name":3147,"fn":3148,"description":3149,"org":3150,"tags":3151,"stars":3161,"repoUrl":3162,"updatedAt":3163},"mcore-build-and-dependency","manage Megatron-LM development environments","Container-based dev environment setup and dependency management for Megatron-LM. Covers acquiring and launching the CI container, uv package management, and updating uv.lock.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3152,3155,3158],{"name":3153,"slug":3154,"type":15},"Containers","containers",{"name":3156,"slug":3157,"type":15},"Deployment","deployment",{"name":3159,"slug":3160,"type":15},"Python","python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":3165,"name":3165,"fn":3166,"description":3167,"org":3168,"tags":3169,"stars":3161,"repoUrl":3162,"updatedAt":3174},"mcore-bump-base-image","update NVIDIA PyTorch base images","Bump the NVIDIA PyTorch base image (`nvcr.io\u002Fnvidia\u002Fpytorch:YY.MM-py3`) used by Megatron-LM CI. Covers the two pin sites (GitHub CI in `docker\u002F.ngc_version.dev` and GitLab CI in `.gitlab\u002Fstages\u002F01.build.yml`), the post-bump CI loop (re-run functional tests, refresh golden values, mark broken tests), and the gotchas that bit PRs",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3170,3173],{"name":3171,"slug":3172,"type":15},"CI\u002FCD","ci-cd",{"name":3156,"slug":3157,"type":15},"2026-07-14T05:25:59.97109",{"slug":3176,"name":3176,"fn":3177,"description":3178,"org":3179,"tags":3180,"stars":3161,"repoUrl":3162,"updatedAt":3186},"mcore-cicd","manage CI\u002FCD pipelines for Megatron-LM","CI\u002FCD reference for Megatron-LM. Covers CI pipeline structure, PR scope labels, triggering internal GitLab CI (which force-pushes the current branch to a pull-request\u002FBRANCH ref — always dry-run and verify the destination first; never run against shared or protected branches), and CI failure investigation.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3181,3182,3183],{"name":3171,"slug":3172,"type":15},{"name":3156,"slug":3157,"type":15},{"name":3184,"slug":3185,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":3188,"name":3188,"fn":3189,"description":3190,"org":3191,"tags":3192,"stars":3161,"repoUrl":3162,"updatedAt":3200},"mcore-create-issue","investigate CI failures and create issues","Investigate a failing GitHub Actions run or job and create a GitHub issue for the failure.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3193,3196,3197],{"name":3194,"slug":3195,"type":15},"Debugging","debugging",{"name":3184,"slug":3185,"type":15},{"name":3198,"slug":3199,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":3202,"name":3202,"fn":3203,"description":3204,"org":3205,"tags":3206,"stars":3161,"repoUrl":3162,"updatedAt":3213},"mcore-linting-and-formatting","lint and format Megatron-LM code","Linting and formatting for Megatron-LM. Covers running autoformat.sh, tools (ruff, black, isort, pylint, mypy), and code style rules.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3207,3210],{"name":3208,"slug":3209,"type":15},"Best Practices","best-practices",{"name":3211,"slug":3212,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":3215,"name":3215,"fn":3216,"description":3217,"org":3218,"tags":3219,"stars":3161,"repoUrl":3162,"updatedAt":3227},"mcore-migrate-gpt-to-hybrid","migrate Megatron-LM models to HybridModel","Migration guide for moving Megatron Core GPTModel checkpoints, model providers, training commands, and layer mappings to HybridModel.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3220,3223,3226],{"name":3221,"slug":3222,"type":15},"Machine Learning","machine-learning",{"name":3224,"slug":3225,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":3229,"name":3229,"fn":3230,"description":3231,"org":3232,"tags":3233,"stars":3161,"repoUrl":3162,"updatedAt":3240},"mcore-onboard-gb200-1node-tests","onboard functional tests for GB200","Onboard 1-node GitHub MR functional tests for GB200 from existing mr-scoped 2-node tests.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3234,3237],{"name":3235,"slug":3236,"type":15},"QA","qa",{"name":3238,"slug":3239,"type":15},"Testing","testing","2026-07-14T05:25:53.673039",{"slug":3242,"name":3242,"fn":3243,"description":3244,"org":3245,"tags":3246,"stars":3161,"repoUrl":3162,"updatedAt":3251},"mcore-run-on-slurm","launch distributed training jobs on SLURM","How to launch distributed Megatron-LM training jobs on a SLURM cluster. Covers a minimal sbatch skeleton, environment-variable setup for torch.distributed.run, CUDA_DEVICE_MAX_CONNECTIONS rules across hardware and parallelism modes, container conventions, monitoring, and per-rank failure diagnosis.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3247,3248],{"name":3156,"slug":3157,"type":15},{"name":3249,"slug":3250,"type":15},"Infrastructure","infrastructure","2026-07-14T05:25:49.362534",{"slug":3253,"name":3253,"fn":3254,"description":3255,"org":3256,"tags":3257,"stars":3161,"repoUrl":3162,"updatedAt":3265},"mcore-split-pr","split pull requests to reduce review load","Split a PR into multiple PRs to reduce the number of required CODEOWNERS reviewer groups.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3258,3261,3262],{"name":3259,"slug":3260,"type":15},"Code Review","code-review",{"name":3184,"slug":3185,"type":15},{"name":3263,"slug":3264,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":3267,"name":3267,"fn":3268,"description":3269,"org":3270,"tags":3271,"stars":3161,"repoUrl":3162,"updatedAt":3274},"mcore-testing","run and manage Megatron-LM tests","Test system for Megatron-LM. Covers test layout, recipe YAML structure, adding and running unit and functional tests, golden values, marker filters, and CI parity.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3272,3273],{"name":3235,"slug":3236,"type":15},{"name":3238,"slug":3239,"type":15},"2026-07-14T05:25:54.928983",{"slug":3276,"name":3276,"fn":3277,"description":3278,"org":3279,"tags":3280,"stars":3161,"repoUrl":3162,"updatedAt":3285},"nightly-sync","manage nightly main-to-dev sync workflows","Domain knowledge for the nightly main-to-dev sync workflow. Covers merge strategy, CI architecture, failure investigation, and known issues.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3281,3284],{"name":3282,"slug":3283,"type":15},"Automation","automation",{"name":3171,"slug":3172,"type":15},"2026-07-30T05:29:03.275638",496,{"items":3288,"total":3382},[3289,3304,3314,3328,3338,3353,3368],{"slug":3290,"name":3290,"fn":3291,"description":3292,"org":3293,"tags":3294,"stars":23,"repoUrl":24,"updatedAt":3303},"accelerated-computing-cudf","accelerate data processing with cuDF","Official NVIDIA-authored guidance for NVIDIA cuDF GPU DataFrames, pandas acceleration, dask-cuDF, ETL, joins, groupby, CSV\u002FParquet I\u002FO, nullable semantics, and multi-GPU DataFrame workloads.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3295,3298,3301,3302],{"name":3296,"slug":3297,"type":15},"Data Analysis","data-analysis",{"name":3299,"slug":3300,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":13,"slug":14,"type":15},"2026-07-14T05:28:43.176466",{"slug":3305,"name":3305,"fn":3306,"description":3307,"org":3308,"tags":3309,"stars":23,"repoUrl":24,"updatedAt":3313},"aiq-deploy","deploy and manage NVIDIA AI-Q infrastructure","Use when asked to install, deploy, run, validate, troubleshoot, or stop NVIDIA AI-Q Blueprint infrastructure.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3310,3311,3312],{"name":3156,"slug":3157,"type":15},{"name":3249,"slug":3250,"type":15},{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":3315,"name":3315,"fn":3316,"description":3317,"org":3318,"tags":3319,"stars":23,"repoUrl":24,"updatedAt":3327},"aiq-research","conduct deep research with AI-Q","Use when asked to run deep research or AI-Q research through a reachable NVIDIA AI-Q Blueprint backend.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3320,3323,3324],{"name":3321,"slug":3322,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":3325,"slug":3326,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":3329,"name":3329,"fn":3330,"description":3331,"org":3332,"tags":3333,"stars":23,"repoUrl":24,"updatedAt":3337},"amc-run-sample-calibration","run AMC sample dataset calibration","Run end-to-end calibration on the shipped sample dataset (sdg_08_2_sample_data_010926.zip) against a running AMC microservice. Use when user says 'test sample dataset', 'run sample calibration', 'verify AMC install', or 'launch and test'.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3334,3335,3336],{"name":3296,"slug":3297,"type":15},{"name":9,"slug":8,"type":15},{"name":3238,"slug":3239,"type":15},"2026-07-17T05:29:03.913266",{"slug":3339,"name":3339,"fn":3340,"description":3341,"org":3342,"tags":3343,"stars":23,"repoUrl":24,"updatedAt":3352},"amc-run-video-calibration","calibrate video datasets with AutoMagicCalib","Calibrate a new dataset from pre-recorded video files via the AutoMagicCalib REST API. Use when user has local MP4s and says 'calibrate my videos', 'run AMC on these videos', or similar. For RTSP\u002Flive streams, use amc-run-rtsp-calibration instead.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3344,3345,3348,3349],{"name":3282,"slug":3283,"type":15},{"name":3346,"slug":3347,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":3350,"slug":3351,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":3354,"name":3354,"fn":3355,"description":3356,"org":3357,"tags":3358,"stars":23,"repoUrl":24,"updatedAt":3367},"amc-setup-calibration-stack","deploy AutoMagicCalib microservice with Docker","Launch AutoMagicCalib microservice and web UI from NGC release images via Docker Compose. Use when user says 'deploy auto calibration', 'launch auto calibration', 'launch AMC', 'start MS+UI', or 'set up auto-magic-calib'. Requires NGC API key.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3359,3360,3363,3364],{"name":3156,"slug":3157,"type":15},{"name":3361,"slug":3362,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":3365,"slug":3366,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":3369,"name":3369,"fn":3370,"description":3371,"org":3372,"tags":3373,"stars":23,"repoUrl":24,"updatedAt":3381},"cudaq-guide","develop quantum applications with CUDA-Q","CUDA-Q onboarding guide for installation, test programs, GPU simulation, QPU hardware, and quantum applications.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3374,3375,3378],{"name":9,"slug":8,"type":15},{"name":3376,"slug":3377,"type":15},"Quantum Computing","quantum-computing",{"name":3379,"slug":3380,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305]