[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-deepstream-import-vision-model":3,"mdc--udibgc-key":34,"related-repo-nvidia-deepstream-import-vision-model":2416,"related-org-nvidia-deepstream-import-vision-model":2520},{"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},"deepstream-import-vision-model","import vision models into DeepStream pipelines","Use this skill to bring any vision model from HuggingFace or NVIDIA NGC into an NVIDIA DeepStream pipeline with end-to-end automation: ONNX download, SafeTensors export, TRT engine build, custom nvinfer bbox parser, multi-stream benchmark, and PDF report. Object detection models only.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},"nvidia","NVIDIA","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fnvidia.png",[12,16,19,20],{"name":13,"slug":14,"type":15},"Deep Learning","deep-learning","tag",{"name":17,"slug":18,"type":15},"Automation","automation",{"name":9,"slug":8,"type":15},{"name":21,"slug":22,"type":15},"Computer Vision","computer-vision",2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-07-17T05:29:08.906831","CC-BY-4.0 AND 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\u002Fdeepstream-import-vision-model","---\nname: deepstream-import-vision-model\ndescription: >\n  Use this skill to bring any vision model from HuggingFace or NVIDIA NGC into\n  an NVIDIA DeepStream pipeline with end-to-end automation: ONNX download,\n  SafeTensors export, TRT engine build, custom nvinfer bbox parser, multi-stream\n  benchmark, and PDF report. Object detection models only.\nlicense: CC-BY-4.0 AND Apache-2.0\nmetadata:\n  author: NVIDIA CORPORATION\n  version: 1.2.2\n---\n\n# DeepStream Import Vision Model\n\nWhen this skill is active, **read the relevant reference document before starting each phase**. Do not rely on memory — reference documents contain exact script paths, bash variable conventions, log filename contracts, and critical parsing rules.\n\n**Current scope:** Object detection models only. Fail fast on classification, segmentation, or other architectures detected in `config.json`.\n\n## Pipeline Overview\n\n| Step | Phase | Reference | What it does |\n|------|-------|-----------|--------------|\n| 1–3 | Model Acquire | [references\u002Fmodel-acquire.md](references\u002Fmodel-acquire.md) | Browse HF\u002FNGC, detect format, download ONNX or export SafeTensors |\n| 4–5 | Engine Build  | [references\u002Fengine-build.md](references\u002Fengine-build.md) | Build dynamic TRT engine, run trtexec BS=1 and BS=MAX_BS |\n| 6–7 | DS Pipeline   | [references\u002Fpipeline-run.md](references\u002Fpipeline-run.md) | Custom bbox parser, nvinfer config, single-stream + multi-stream benchmarks |\n| 8   | Report        | [references\u002Freport-generation.md](references\u002Freport-generation.md) | 5 charts, HTML, PDF benchmark report |\n\nRun the full pipeline autonomously without pausing for confirmation at each step.\n\n## Pre-flight Checks\n\nRun before starting:\n\n```bash\n# 1. GPU and drivers\nnvidia-smi\n\n# 2. TensorRT version match (must match between builder and DS runtime)\ntrtexec 2>&1 | head -3\ndpkg -l | grep libnvinfer-bin\n\n# 3. Shared Python venv — create once, reuse across all models\nmkdir -p build\nVENV=build\u002F.venv_optimum\nif [ ! -x \"$VENV\u002Fbin\u002Fpython3\" ]; then\n  python3 -m venv \"$VENV\"\n  \"$VENV\u002Fbin\u002Fpip\" install --upgrade pip -q\n  \"$VENV\u002Fbin\u002Fpip\" install \"optimum[exporters]>=1.20,\u003C2.0\" \"torch\u003C2.12\" \\\n    transformers onnxruntime matplotlib numpy markdown -q\nfi\n\n# 4. System tools\nwhich wkhtmltopdf || apt-get install -y wkhtmltopdf\nwhich mediainfo    || apt-get install -y mediainfo\nwhich deepstream-app  # required for KITTI dump (Step 6g) and benchmark perf-measurement (Step 7c); shipped with DeepStream SDK\n\n# 5. Sample video — only check default path when user has not provided a custom DS_VIDEO\nif [ -z \"$DS_VIDEO\" ]; then\n  [ -f \u002Fopt\u002Fnvidia\u002Fdeepstream\u002Fdeepstream\u002Fsamples\u002Fstreams\u002Fsample_720p.mp4 ] || \\\n    echo \"WARNING: sample_720p.mp4 not found. Install DeepStream samples or set DS_VIDEO=\u002Fpath\u002Fto\u002Fyour.mp4\"\nfi\n```\n\n## Mandatory Output Structure\n\nCreate once `MODEL_NAME` is known (Step 1). Never dump files flat.\n\n```\nmodels\u002F{model_name}\u002F\n  model\u002F           \u003C- ONNX file(s)\n  parser\u002F          \u003C- .cpp, Makefile, .so\n  config\u002F          \u003C- nvinfer config, ds-app config, labels.txt\n  scripts\u002F         \u003C- run helper scripts\n  benchmarks\u002F\n    engines\u002F       \u003C- _dynamic_b{MAX_BS}.engine, timing.cache, build logs\n    b1\u002F            \u003C- trtexec BS=1 log\n    b{MAX_BS}\u002F     \u003C- trtexec BS=MAX_BS log\n    ds\u002F            \u003C- DS benchmark logs\n  reports\u002F         \u003C- benchmark_report.md, .html, .pdf, benchmark_data.json\n    charts\u002F        \u003C- chart_*.png (5 charts)\n  samples\u002F         \u003C- output .mp4 or .ogv (theoraenc fallback), test frames\n    kitti_output\u002F  \u003C- KITTI detection .txt files\n```\n\n```bash\nmkdir -p models\u002F$MODEL_NAME\u002F{model,parser,config,scripts,benchmarks\u002Fengines,benchmarks\u002Fds,reports\u002Fcharts,samples\u002Fkitti_output}\n```\n\n## Critical Rules\n\n1. **Engine naming** — always `{model}_dynamic_b{MAX_BS}.engine`. Never bare `model_dynamic.engine`.\n2. **batch_size == num_streams** — in DS runs, `batch-size` and stream count are always equal.\n3. **Log filenames are fixed** — `trtexec_b1.log`, `trtexec_b${MAX_BS}.log`, `ds_s${N}_run1.log`, `ds_s${N}_run2.log`. No timestamps. Report generation reads exact paths.\n4. **Parser zero-init** — always `NvDsInferObjectDetectionInfo obj = {};`. Required for DeepStream OBB support; bare `obj;` leaves `rotation_angle` uninitialized, causing tilted bounding boxes.\n5. **KITTI validation gate** — do NOT proceed to Step 7 if KITTI frame count is zero or detection rate \u003C 90%.\n6. **Shared venv** — `build\u002F.venv_optimum` reused across all models. Never create per-model venvs.\n7. **trtexec `--noDataTransfers`** — GPU-only compute matches DeepStream's GPU-to-GPU data flow.\n8. **Report HTML+PDF** — always use `skills\u002Fdeepstream-import-vision-model\u002Fscripts\u002Freport\u002Fmd-to-html-pdf.py`. Never write a custom HTML generator or call `wkhtmltopdf` directly.\n9. **Object detection only** — reject non-detection architectures from `config.json` before building anything.\n10. **Encoder fallback (MANDATORY)** — `x264enc` and `openh264enc` are **prohibited**. On NVENC-unavailable systems, use `theoraenc + oggmux` (LGPL; ships in gst-plugins-base; output is `.ogv`). If `theoraenc`\u002F`oggmux` are absent, skip video creation (`DS_SINGLE_STREAM_MODE=skipped`). Report which mode was used: `nvv4l2h264enc` \u002F `theoraenc-fallback` \u002F `skipped`.\n11. **Video source (MANDATORY)** — default is always `sample_720p.mp4` (1280×720). Never autonomously substitute `sample_1080p_h264.mp4` or any other file. Only use a different video when the user explicitly provides a path (via `DS_VIDEO` env var or script argument).\n\n## Pipeline Timing\n\nWrap every step:\n\n```bash\nSTEP_START=$(date +%s.%N)\n# ... step commands ...\nSTEP_END=$(date +%s.%N)\nSTEP_DURATION=$(echo \"$STEP_END - $STEP_START\" | bc)\necho \"[Step N] completed in ${STEP_DURATION}s\"\n```\n\nTrack `PIPELINE_START` (before Step 1) and `PIPELINE_END` (after Step 8). Report all durations in the benchmark report.\n\n## Report Output (MANDATORY — all 3 formats)\n\n1. `benchmark_report.md` — markdown source (12 mandatory sections)\n2. `benchmark_report.html` — styled HTML (charts base64-inlined, no local file access)\n3. `benchmark_report_{model_name}.pdf` — via `md-to-html-pdf.py`; verify charts are embedded by counting `data:image\u002Fpng` occurrences in the HTML output: `grep -o 'data:image\u002Fpng' benchmark_report.html | wc -l` should equal 5\n\nRun charts and report scripts with the shared venv active: `source build\u002F.venv_optimum\u002Fbin\u002Factivate`.\n\n## Reference Documents\n\n**IMPORTANT**: Read the relevant reference before starting each phase. Do NOT generate code from memory.\n\n| Document | Use When |\n|----------|----------|\n| [references\u002Fmodel-acquire.md](references\u002Fmodel-acquire.md) | Steps 1–3: HF\u002FNGC URL parsing, format detection, ONNX download, SafeTensors export, label extraction |\n| [references\u002Fengine-build.md](references\u002Fengine-build.md) | Steps 4–5: trtexec engine build, benchmarks, PEAK_GPU_STREAMS derivation, iterative scaling |\n| [references\u002Fpipeline-run.md](references\u002Fpipeline-run.md) | Steps 6–7: custom bbox parser, nvinfer config, single-stream validation, KITTI dump, multi-stream benchmark |\n| [references\u002Freport-generation.md](references\u002Freport-generation.md) | Step 8: benchmark_data.json, 5 charts, 12-section markdown report, HTML + PDF |\n\n## Scripts\n\nLocated in `scripts\u002F`.\n\n| Script | Phase | Purpose |\n|--------|-------|---------|\n| `model\u002Fhf-list-files.sh` | 1–3 | List HuggingFace repo files |\n| `model\u002Fhf-download-config.sh` | 1–3 | Download config.json from HF |\n| `model\u002Fngc-list-files.sh` | 1–3 | List NGC model files |\n| `model\u002Fngc-download.sh` | 1–3 | Download NGC model archive |\n| `model\u002Fsafetensors-to-onnx.sh` | 1–3 | Export SafeTensors → ONNX via optimum-cli |\n| `model\u002Finspect-onnx.py` | 1–5 | Inspect ONNX input\u002Foutput shapes |\n| `model\u002Fmake-static-batch-onnx.py` | 4–5 | Bake batch dim into ONNX |\n| `model\u002Fcleanup.sh` | Any | Remove staging dirs, preserve shared venv |\n| `engine\u002Fbenchmark-trtexec.sh` | 4–5 | Run trtexec with standard flags |\n| `deepstream\u002Fds-single-stream.sh` | 6–7 | Single-stream visual validation (NVENC primary; theoraenc+oggmux fallback; skip if neither) |\n| `deepstream\u002Fds-sweep.sh` | 6–7 | 2-phase batch size sweep |\n| `deepstream\u002Fbenchmark-ds.sh` | 6–7 | Fixed-stream DS benchmark |\n| `deepstream\u002Fds-kitti-dump.sh` | 6–7 | KITTI detection dump via deepstream-app |\n| `deepstream\u002Fds-perf-run.sh` | 7 | Step 7c two-run benchmark — wraps `deepstream-app` with `enable-perf-measurement=1`, writes fixed-name log for the report parser |\n| `deepstream\u002Fextract-frame.sh` | 6–7 | Extract sample frames from output video (`.mp4` NVENC path or `.ogv` theoraenc fallback) |\n| `report\u002Fgenerate-benchmark-charts.py` | 8 | Generate 5 benchmark PNG charts |\n| `report\u002Fmd-to-html-pdf.py` | 8 | Markdown → styled HTML → PDF (canonical benchmark report path) |\n| `report\u002Fmd-to-pdf.sh` | Any | Markdown → PDF via pandoc\u002Fpdflatex — for design docs and references only, NOT for benchmark reports (use md-to-html-pdf.py for those) |\n| `report\u002Freport-style.css` | 8 | CSS for HTML report |\n| `report\u002Frender-mermaid-for-pdf.py` | 8 | Mermaid diagram → PNG |\n| `report\u002Fmermaid-puppeteer.json` | 8 | Vetted Puppeteer config for Mermaid (sandboxed; non-root) |\n| `report\u002Fmermaid-puppeteer-root.json` | 8 | Vetted Puppeteer config for Mermaid (used when running as root) |\n\n## Quick Error Reference\n\n| Error | Fix |\n|-------|-----|\n| Tilted\u002Fdiagonal bounding boxes | Parser struct not zero-initialized — use `NvDsInferObjectDetectionInfo obj = {};` |\n| Zero KITTI files | `gie-kitti-output-dir` not read by nvinfer — use `ds-kitti-dump.sh` (wraps `deepstream-app`) |\n| Engine rebuilds every DS run | `model-engine-file` path wrong — check relative path from `config\u002F` dir |\n| `setDimensions` negative dims | Add `infer-dims=3;H;W` to nvinfer config for dynamic ONNX models |\n| `--memPoolSize` workspace 0.03 MiB | Use `M` suffix not `MiB` — e.g. `--memPoolSize=workspace:32768M` |\n| ForeignNode build failure (DETR) | Use dynamo export path or run `onnxsim` — see references\u002Fengine-build.md |\n| Zero detections | Wrong `net-scale-factor` — check model family table in references\u002Fpipeline-run.md |\n| `No module named 'pyservicemaker'` | Install into venv: `pip install \u002Fopt\u002Fnvidia\u002Fdeepstream\u002F...\u002Fpyservicemaker*.whl` |\n\n\u003C!-- Signing refresh marker. -->\n",{"data":35,"body":39},{"name":4,"description":6,"license":26,"metadata":36},{"author":37,"version":38},"NVIDIA CORPORATION","1.2.2",{"type":40,"children":41},"root",[42,50,64,83,90,233,238,244,249,860,866,879,889,922,928,1263,1269,1274,1433,1454,1460,1520,1532,1538,1548,1630,1636,1648,2170,2176,2410],{"type":43,"tag":44,"props":45,"children":46},"element","h1",{"id":4},[47],{"type":48,"value":49},"text","DeepStream Import Vision Model",{"type":43,"tag":51,"props":52,"children":53},"p",{},[54,56,62],{"type":48,"value":55},"When this skill is active, ",{"type":43,"tag":57,"props":58,"children":59},"strong",{},[60],{"type":48,"value":61},"read the relevant reference document before starting each phase",{"type":48,"value":63},". Do not rely on memory — reference documents contain exact script paths, bash variable conventions, log filename contracts, and critical parsing rules.",{"type":43,"tag":51,"props":65,"children":66},{},[67,72,74,81],{"type":43,"tag":57,"props":68,"children":69},{},[70],{"type":48,"value":71},"Current scope:",{"type":48,"value":73}," Object detection models only. Fail fast on classification, segmentation, or other architectures detected in ",{"type":43,"tag":75,"props":76,"children":78},"code",{"className":77},[],[79],{"type":48,"value":80},"config.json",{"type":48,"value":82},".",{"type":43,"tag":84,"props":85,"children":87},"h2",{"id":86},"pipeline-overview",[88],{"type":48,"value":89},"Pipeline Overview",{"type":43,"tag":91,"props":92,"children":93},"table",{},[94,123],{"type":43,"tag":95,"props":96,"children":97},"thead",{},[98],{"type":43,"tag":99,"props":100,"children":101},"tr",{},[102,108,113,118],{"type":43,"tag":103,"props":104,"children":105},"th",{},[106],{"type":48,"value":107},"Step",{"type":43,"tag":103,"props":109,"children":110},{},[111],{"type":48,"value":112},"Phase",{"type":43,"tag":103,"props":114,"children":115},{},[116],{"type":48,"value":117},"Reference",{"type":43,"tag":103,"props":119,"children":120},{},[121],{"type":48,"value":122},"What it does",{"type":43,"tag":124,"props":125,"children":126},"tbody",{},[127,155,181,207],{"type":43,"tag":99,"props":128,"children":129},{},[130,136,141,150],{"type":43,"tag":131,"props":132,"children":133},"td",{},[134],{"type":48,"value":135},"1–3",{"type":43,"tag":131,"props":137,"children":138},{},[139],{"type":48,"value":140},"Model Acquire",{"type":43,"tag":131,"props":142,"children":143},{},[144],{"type":43,"tag":145,"props":146,"children":148},"a",{"href":147},"references\u002Fmodel-acquire.md",[149],{"type":48,"value":147},{"type":43,"tag":131,"props":151,"children":152},{},[153],{"type":48,"value":154},"Browse HF\u002FNGC, detect format, download ONNX or export SafeTensors",{"type":43,"tag":99,"props":156,"children":157},{},[158,163,168,176],{"type":43,"tag":131,"props":159,"children":160},{},[161],{"type":48,"value":162},"4–5",{"type":43,"tag":131,"props":164,"children":165},{},[166],{"type":48,"value":167},"Engine Build",{"type":43,"tag":131,"props":169,"children":170},{},[171],{"type":43,"tag":145,"props":172,"children":174},{"href":173},"references\u002Fengine-build.md",[175],{"type":48,"value":173},{"type":43,"tag":131,"props":177,"children":178},{},[179],{"type":48,"value":180},"Build dynamic TRT engine, run trtexec BS=1 and BS=MAX_BS",{"type":43,"tag":99,"props":182,"children":183},{},[184,189,194,202],{"type":43,"tag":131,"props":185,"children":186},{},[187],{"type":48,"value":188},"6–7",{"type":43,"tag":131,"props":190,"children":191},{},[192],{"type":48,"value":193},"DS Pipeline",{"type":43,"tag":131,"props":195,"children":196},{},[197],{"type":43,"tag":145,"props":198,"children":200},{"href":199},"references\u002Fpipeline-run.md",[201],{"type":48,"value":199},{"type":43,"tag":131,"props":203,"children":204},{},[205],{"type":48,"value":206},"Custom bbox parser, nvinfer config, single-stream + multi-stream benchmarks",{"type":43,"tag":99,"props":208,"children":209},{},[210,215,220,228],{"type":43,"tag":131,"props":211,"children":212},{},[213],{"type":48,"value":214},"8",{"type":43,"tag":131,"props":216,"children":217},{},[218],{"type":48,"value":219},"Report",{"type":43,"tag":131,"props":221,"children":222},{},[223],{"type":43,"tag":145,"props":224,"children":226},{"href":225},"references\u002Freport-generation.md",[227],{"type":48,"value":225},{"type":43,"tag":131,"props":229,"children":230},{},[231],{"type":48,"value":232},"5 charts, HTML, PDF benchmark report",{"type":43,"tag":51,"props":234,"children":235},{},[236],{"type":48,"value":237},"Run the full pipeline autonomously without pausing for confirmation at each step.",{"type":43,"tag":84,"props":239,"children":241},{"id":240},"pre-flight-checks",[242],{"type":48,"value":243},"Pre-flight Checks",{"type":43,"tag":51,"props":245,"children":246},{},[247],{"type":48,"value":248},"Run before starting:",{"type":43,"tag":250,"props":251,"children":256},"pre",{"className":252,"code":253,"language":254,"meta":255,"style":255},"language-bash shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","# 1. GPU and drivers\nnvidia-smi\n\n# 2. TensorRT version match (must match between builder and DS runtime)\ntrtexec 2>&1 | head -3\ndpkg -l | grep libnvinfer-bin\n\n# 3. Shared Python venv — create once, reuse across all models\nmkdir -p build\nVENV=build\u002F.venv_optimum\nif [ ! -x \"$VENV\u002Fbin\u002Fpython3\" ]; then\n  python3 -m venv \"$VENV\"\n  \"$VENV\u002Fbin\u002Fpip\" install --upgrade pip -q\n  \"$VENV\u002Fbin\u002Fpip\" install \"optimum[exporters]>=1.20,\u003C2.0\" \"torch\u003C2.12\" \\\n    transformers onnxruntime matplotlib numpy markdown -q\nfi\n\n# 4. System tools\nwhich wkhtmltopdf || apt-get install -y wkhtmltopdf\nwhich mediainfo    || apt-get install -y mediainfo\nwhich deepstream-app  # required for KITTI dump (Step 6g) and benchmark perf-measurement (Step 7c); shipped with DeepStream SDK\n\n# 5. Sample video — only check default path when user has not provided a custom DS_VIDEO\nif [ -z \"$DS_VIDEO\" ]; then\n  [ -f \u002Fopt\u002Fnvidia\u002Fdeepstream\u002Fdeepstream\u002Fsamples\u002Fstreams\u002Fsample_720p.mp4 ] || \\\n    echo \"WARNING: sample_720p.mp4 not found. Install DeepStream samples or set DS_VIDEO=\u002Fpath\u002Fto\u002Fyour.mp4\"\nfi\n","bash","",[257],{"type":43,"tag":75,"props":258,"children":259},{"__ignoreMap":255},[260,272,282,292,301,332,360,368,377,396,416,471,503,541,592,625,634,642,651,690,725,743,751,760,798,830,852],{"type":43,"tag":261,"props":262,"children":265},"span",{"class":263,"line":264},"line",1,[266],{"type":43,"tag":261,"props":267,"children":269},{"style":268},"--shiki-light:#90A4AE;--shiki-light-font-style:italic;--shiki-default:#546E7A;--shiki-default-font-style:italic;--shiki-dark:#676E95;--shiki-dark-font-style:italic",[270],{"type":48,"value":271},"# 1. GPU and drivers\n",{"type":43,"tag":261,"props":273,"children":275},{"class":263,"line":274},2,[276],{"type":43,"tag":261,"props":277,"children":279},{"style":278},"--shiki-light:#E2931D;--shiki-default:#FFCB6B;--shiki-dark:#FFCB6B",[280],{"type":48,"value":281},"nvidia-smi\n",{"type":43,"tag":261,"props":283,"children":285},{"class":263,"line":284},3,[286],{"type":43,"tag":261,"props":287,"children":289},{"emptyLinePlaceholder":288},true,[290],{"type":48,"value":291},"\n",{"type":43,"tag":261,"props":293,"children":295},{"class":263,"line":294},4,[296],{"type":43,"tag":261,"props":297,"children":298},{"style":268},[299],{"type":48,"value":300},"# 2. TensorRT version match (must match between builder and DS runtime)\n",{"type":43,"tag":261,"props":302,"children":304},{"class":263,"line":303},5,[305,310,316,321,326],{"type":43,"tag":261,"props":306,"children":307},{"style":278},[308],{"type":48,"value":309},"trtexec",{"type":43,"tag":261,"props":311,"children":313},{"style":312},"--shiki-light:#39ADB5;--shiki-default:#89DDFF;--shiki-dark:#89DDFF",[314],{"type":48,"value":315}," 2>&1",{"type":43,"tag":261,"props":317,"children":318},{"style":312},[319],{"type":48,"value":320}," |",{"type":43,"tag":261,"props":322,"children":323},{"style":278},[324],{"type":48,"value":325}," head",{"type":43,"tag":261,"props":327,"children":329},{"style":328},"--shiki-light:#91B859;--shiki-default:#C3E88D;--shiki-dark:#C3E88D",[330],{"type":48,"value":331}," -3\n",{"type":43,"tag":261,"props":333,"children":335},{"class":263,"line":334},6,[336,341,346,350,355],{"type":43,"tag":261,"props":337,"children":338},{"style":278},[339],{"type":48,"value":340},"dpkg",{"type":43,"tag":261,"props":342,"children":343},{"style":328},[344],{"type":48,"value":345}," -l",{"type":43,"tag":261,"props":347,"children":348},{"style":312},[349],{"type":48,"value":320},{"type":43,"tag":261,"props":351,"children":352},{"style":278},[353],{"type":48,"value":354}," grep",{"type":43,"tag":261,"props":356,"children":357},{"style":328},[358],{"type":48,"value":359}," libnvinfer-bin\n",{"type":43,"tag":261,"props":361,"children":363},{"class":263,"line":362},7,[364],{"type":43,"tag":261,"props":365,"children":366},{"emptyLinePlaceholder":288},[367],{"type":48,"value":291},{"type":43,"tag":261,"props":369,"children":371},{"class":263,"line":370},8,[372],{"type":43,"tag":261,"props":373,"children":374},{"style":268},[375],{"type":48,"value":376},"# 3. Shared Python venv — create once, reuse across all models\n",{"type":43,"tag":261,"props":378,"children":380},{"class":263,"line":379},9,[381,386,391],{"type":43,"tag":261,"props":382,"children":383},{"style":278},[384],{"type":48,"value":385},"mkdir",{"type":43,"tag":261,"props":387,"children":388},{"style":328},[389],{"type":48,"value":390}," -p",{"type":43,"tag":261,"props":392,"children":393},{"style":328},[394],{"type":48,"value":395}," build\n",{"type":43,"tag":261,"props":397,"children":399},{"class":263,"line":398},10,[400,406,411],{"type":43,"tag":261,"props":401,"children":403},{"style":402},"--shiki-light:#90A4AE;--shiki-default:#EEFFFF;--shiki-dark:#BABED8",[404],{"type":48,"value":405},"VENV",{"type":43,"tag":261,"props":407,"children":408},{"style":312},[409],{"type":48,"value":410},"=",{"type":43,"tag":261,"props":412,"children":413},{"style":328},[414],{"type":48,"value":415},"build\u002F.venv_optimum\n",{"type":43,"tag":261,"props":417,"children":419},{"class":263,"line":418},11,[420,426,431,436,441,446,451,456,461,466],{"type":43,"tag":261,"props":421,"children":423},{"style":422},"--shiki-light:#39ADB5;--shiki-light-font-style:italic;--shiki-default:#89DDFF;--shiki-default-font-style:italic;--shiki-dark:#89DDFF;--shiki-dark-font-style:italic",[424],{"type":48,"value":425},"if",{"type":43,"tag":261,"props":427,"children":428},{"style":312},[429],{"type":48,"value":430}," [",{"type":43,"tag":261,"props":432,"children":433},{"style":312},[434],{"type":48,"value":435}," !",{"type":43,"tag":261,"props":437,"children":438},{"style":312},[439],{"type":48,"value":440}," -x",{"type":43,"tag":261,"props":442,"children":443},{"style":312},[444],{"type":48,"value":445}," \"",{"type":43,"tag":261,"props":447,"children":448},{"style":402},[449],{"type":48,"value":450},"$VENV",{"type":43,"tag":261,"props":452,"children":453},{"style":328},[454],{"type":48,"value":455},"\u002Fbin\u002Fpython3",{"type":43,"tag":261,"props":457,"children":458},{"style":312},[459],{"type":48,"value":460},"\"",{"type":43,"tag":261,"props":462,"children":463},{"style":312},[464],{"type":48,"value":465}," ];",{"type":43,"tag":261,"props":467,"children":468},{"style":422},[469],{"type":48,"value":470}," then\n",{"type":43,"tag":261,"props":472,"children":474},{"class":263,"line":473},12,[475,480,485,490,494,498],{"type":43,"tag":261,"props":476,"children":477},{"style":278},[478],{"type":48,"value":479},"  python3",{"type":43,"tag":261,"props":481,"children":482},{"style":328},[483],{"type":48,"value":484}," -m",{"type":43,"tag":261,"props":486,"children":487},{"style":328},[488],{"type":48,"value":489}," venv",{"type":43,"tag":261,"props":491,"children":492},{"style":312},[493],{"type":48,"value":445},{"type":43,"tag":261,"props":495,"children":496},{"style":402},[497],{"type":48,"value":450},{"type":43,"tag":261,"props":499,"children":500},{"style":312},[501],{"type":48,"value":502},"\"\n",{"type":43,"tag":261,"props":504,"children":506},{"class":263,"line":505},13,[507,512,516,521,526,531,536],{"type":43,"tag":261,"props":508,"children":509},{"style":278},[510],{"type":48,"value":511},"  \"",{"type":43,"tag":261,"props":513,"children":514},{"style":402},[515],{"type":48,"value":450},{"type":43,"tag":261,"props":517,"children":518},{"style":278},[519],{"type":48,"value":520},"\u002Fbin\u002Fpip\"",{"type":43,"tag":261,"props":522,"children":523},{"style":328},[524],{"type":48,"value":525}," install",{"type":43,"tag":261,"props":527,"children":528},{"style":328},[529],{"type":48,"value":530}," --upgrade",{"type":43,"tag":261,"props":532,"children":533},{"style":328},[534],{"type":48,"value":535}," pip",{"type":43,"tag":261,"props":537,"children":538},{"style":328},[539],{"type":48,"value":540}," -q\n",{"type":43,"tag":261,"props":542,"children":544},{"class":263,"line":543},14,[545,549,553,557,561,565,570,574,578,583,587],{"type":43,"tag":261,"props":546,"children":547},{"style":278},[548],{"type":48,"value":511},{"type":43,"tag":261,"props":550,"children":551},{"style":402},[552],{"type":48,"value":450},{"type":43,"tag":261,"props":554,"children":555},{"style":278},[556],{"type":48,"value":520},{"type":43,"tag":261,"props":558,"children":559},{"style":328},[560],{"type":48,"value":525},{"type":43,"tag":261,"props":562,"children":563},{"style":312},[564],{"type":48,"value":445},{"type":43,"tag":261,"props":566,"children":567},{"style":328},[568],{"type":48,"value":569},"optimum[exporters]>=1.20,\u003C2.0",{"type":43,"tag":261,"props":571,"children":572},{"style":312},[573],{"type":48,"value":460},{"type":43,"tag":261,"props":575,"children":576},{"style":312},[577],{"type":48,"value":445},{"type":43,"tag":261,"props":579,"children":580},{"style":328},[581],{"type":48,"value":582},"torch\u003C2.12",{"type":43,"tag":261,"props":584,"children":585},{"style":312},[586],{"type":48,"value":460},{"type":43,"tag":261,"props":588,"children":589},{"style":402},[590],{"type":48,"value":591}," \\\n",{"type":43,"tag":261,"props":593,"children":595},{"class":263,"line":594},15,[596,601,606,611,616,621],{"type":43,"tag":261,"props":597,"children":598},{"style":328},[599],{"type":48,"value":600},"    transformers",{"type":43,"tag":261,"props":602,"children":603},{"style":328},[604],{"type":48,"value":605}," onnxruntime",{"type":43,"tag":261,"props":607,"children":608},{"style":328},[609],{"type":48,"value":610}," matplotlib",{"type":43,"tag":261,"props":612,"children":613},{"style":328},[614],{"type":48,"value":615}," numpy",{"type":43,"tag":261,"props":617,"children":618},{"style":328},[619],{"type":48,"value":620}," markdown",{"type":43,"tag":261,"props":622,"children":623},{"style":328},[624],{"type":48,"value":540},{"type":43,"tag":261,"props":626,"children":628},{"class":263,"line":627},16,[629],{"type":43,"tag":261,"props":630,"children":631},{"style":422},[632],{"type":48,"value":633},"fi\n",{"type":43,"tag":261,"props":635,"children":637},{"class":263,"line":636},17,[638],{"type":43,"tag":261,"props":639,"children":640},{"emptyLinePlaceholder":288},[641],{"type":48,"value":291},{"type":43,"tag":261,"props":643,"children":645},{"class":263,"line":644},18,[646],{"type":43,"tag":261,"props":647,"children":648},{"style":268},[649],{"type":48,"value":650},"# 4. System tools\n",{"type":43,"tag":261,"props":652,"children":654},{"class":263,"line":653},19,[655,661,666,671,676,680,685],{"type":43,"tag":261,"props":656,"children":658},{"style":657},"--shiki-light:#6182B8;--shiki-default:#82AAFF;--shiki-dark:#82AAFF",[659],{"type":48,"value":660},"which",{"type":43,"tag":261,"props":662,"children":663},{"style":328},[664],{"type":48,"value":665}," wkhtmltopdf",{"type":43,"tag":261,"props":667,"children":668},{"style":312},[669],{"type":48,"value":670}," ||",{"type":43,"tag":261,"props":672,"children":673},{"style":278},[674],{"type":48,"value":675}," apt-get",{"type":43,"tag":261,"props":677,"children":678},{"style":328},[679],{"type":48,"value":525},{"type":43,"tag":261,"props":681,"children":682},{"style":328},[683],{"type":48,"value":684}," -y",{"type":43,"tag":261,"props":686,"children":687},{"style":328},[688],{"type":48,"value":689}," wkhtmltopdf\n",{"type":43,"tag":261,"props":691,"children":693},{"class":263,"line":692},20,[694,698,703,708,712,716,720],{"type":43,"tag":261,"props":695,"children":696},{"style":657},[697],{"type":48,"value":660},{"type":43,"tag":261,"props":699,"children":700},{"style":328},[701],{"type":48,"value":702}," mediainfo",{"type":43,"tag":261,"props":704,"children":705},{"style":312},[706],{"type":48,"value":707},"    ||",{"type":43,"tag":261,"props":709,"children":710},{"style":278},[711],{"type":48,"value":675},{"type":43,"tag":261,"props":713,"children":714},{"style":328},[715],{"type":48,"value":525},{"type":43,"tag":261,"props":717,"children":718},{"style":328},[719],{"type":48,"value":684},{"type":43,"tag":261,"props":721,"children":722},{"style":328},[723],{"type":48,"value":724}," mediainfo\n",{"type":43,"tag":261,"props":726,"children":728},{"class":263,"line":727},21,[729,733,738],{"type":43,"tag":261,"props":730,"children":731},{"style":657},[732],{"type":48,"value":660},{"type":43,"tag":261,"props":734,"children":735},{"style":328},[736],{"type":48,"value":737}," deepstream-app",{"type":43,"tag":261,"props":739,"children":740},{"style":268},[741],{"type":48,"value":742},"  # required for KITTI dump (Step 6g) and benchmark perf-measurement (Step 7c); shipped with DeepStream SDK\n",{"type":43,"tag":261,"props":744,"children":746},{"class":263,"line":745},22,[747],{"type":43,"tag":261,"props":748,"children":749},{"emptyLinePlaceholder":288},[750],{"type":48,"value":291},{"type":43,"tag":261,"props":752,"children":754},{"class":263,"line":753},23,[755],{"type":43,"tag":261,"props":756,"children":757},{"style":268},[758],{"type":48,"value":759},"# 5. Sample video — only check default path when user has not provided a custom DS_VIDEO\n",{"type":43,"tag":261,"props":761,"children":763},{"class":263,"line":762},24,[764,768,772,777,781,786,790,794],{"type":43,"tag":261,"props":765,"children":766},{"style":422},[767],{"type":48,"value":425},{"type":43,"tag":261,"props":769,"children":770},{"style":312},[771],{"type":48,"value":430},{"type":43,"tag":261,"props":773,"children":774},{"style":312},[775],{"type":48,"value":776}," -z",{"type":43,"tag":261,"props":778,"children":779},{"style":312},[780],{"type":48,"value":445},{"type":43,"tag":261,"props":782,"children":783},{"style":402},[784],{"type":48,"value":785},"$DS_VIDEO",{"type":43,"tag":261,"props":787,"children":788},{"style":312},[789],{"type":48,"value":460},{"type":43,"tag":261,"props":791,"children":792},{"style":312},[793],{"type":48,"value":465},{"type":43,"tag":261,"props":795,"children":796},{"style":422},[797],{"type":48,"value":470},{"type":43,"tag":261,"props":799,"children":801},{"class":263,"line":800},25,[802,807,812,817,822,826],{"type":43,"tag":261,"props":803,"children":804},{"style":312},[805],{"type":48,"value":806},"  [",{"type":43,"tag":261,"props":808,"children":809},{"style":312},[810],{"type":48,"value":811}," -f",{"type":43,"tag":261,"props":813,"children":814},{"style":402},[815],{"type":48,"value":816}," \u002Fopt\u002Fnvidia\u002Fdeepstream\u002Fdeepstream\u002Fsamples\u002Fstreams\u002Fsample_720p.mp4 ",{"type":43,"tag":261,"props":818,"children":819},{"style":312},[820],{"type":48,"value":821},"]",{"type":43,"tag":261,"props":823,"children":824},{"style":312},[825],{"type":48,"value":670},{"type":43,"tag":261,"props":827,"children":828},{"style":402},[829],{"type":48,"value":591},{"type":43,"tag":261,"props":831,"children":833},{"class":263,"line":832},26,[834,839,843,848],{"type":43,"tag":261,"props":835,"children":836},{"style":657},[837],{"type":48,"value":838},"    echo",{"type":43,"tag":261,"props":840,"children":841},{"style":312},[842],{"type":48,"value":445},{"type":43,"tag":261,"props":844,"children":845},{"style":328},[846],{"type":48,"value":847},"WARNING: sample_720p.mp4 not found. Install DeepStream samples or set DS_VIDEO=\u002Fpath\u002Fto\u002Fyour.mp4",{"type":43,"tag":261,"props":849,"children":850},{"style":312},[851],{"type":48,"value":502},{"type":43,"tag":261,"props":853,"children":855},{"class":263,"line":854},27,[856],{"type":43,"tag":261,"props":857,"children":858},{"style":422},[859],{"type":48,"value":633},{"type":43,"tag":84,"props":861,"children":863},{"id":862},"mandatory-output-structure",[864],{"type":48,"value":865},"Mandatory Output Structure",{"type":43,"tag":51,"props":867,"children":868},{},[869,871,877],{"type":48,"value":870},"Create once ",{"type":43,"tag":75,"props":872,"children":874},{"className":873},[],[875],{"type":48,"value":876},"MODEL_NAME",{"type":48,"value":878}," is known (Step 1). Never dump files flat.",{"type":43,"tag":250,"props":880,"children":884},{"className":881,"code":883,"language":48},[882],"language-text","models\u002F{model_name}\u002F\n  model\u002F           \u003C- ONNX file(s)\n  parser\u002F          \u003C- .cpp, Makefile, .so\n  config\u002F          \u003C- nvinfer config, ds-app config, labels.txt\n  scripts\u002F         \u003C- run helper scripts\n  benchmarks\u002F\n    engines\u002F       \u003C- _dynamic_b{MAX_BS}.engine, timing.cache, build logs\n    b1\u002F            \u003C- trtexec BS=1 log\n    b{MAX_BS}\u002F     \u003C- trtexec BS=MAX_BS log\n    ds\u002F            \u003C- DS benchmark logs\n  reports\u002F         \u003C- benchmark_report.md, .html, .pdf, benchmark_data.json\n    charts\u002F        \u003C- chart_*.png (5 charts)\n  samples\u002F         \u003C- output .mp4 or .ogv (theoraenc fallback), test frames\n    kitti_output\u002F  \u003C- KITTI detection .txt files\n",[885],{"type":43,"tag":75,"props":886,"children":887},{"__ignoreMap":255},[888],{"type":48,"value":883},{"type":43,"tag":250,"props":890,"children":892},{"className":252,"code":891,"language":254,"meta":255,"style":255},"mkdir -p models\u002F$MODEL_NAME\u002F{model,parser,config,scripts,benchmarks\u002Fengines,benchmarks\u002Fds,reports\u002Fcharts,samples\u002Fkitti_output}\n",[893],{"type":43,"tag":75,"props":894,"children":895},{"__ignoreMap":255},[896],{"type":43,"tag":261,"props":897,"children":898},{"class":263,"line":264},[899,903,907,912,917],{"type":43,"tag":261,"props":900,"children":901},{"style":278},[902],{"type":48,"value":385},{"type":43,"tag":261,"props":904,"children":905},{"style":328},[906],{"type":48,"value":390},{"type":43,"tag":261,"props":908,"children":909},{"style":328},[910],{"type":48,"value":911}," models\u002F",{"type":43,"tag":261,"props":913,"children":914},{"style":402},[915],{"type":48,"value":916},"$MODEL_NAME",{"type":43,"tag":261,"props":918,"children":919},{"style":328},[920],{"type":48,"value":921},"\u002F{model,parser,config,scripts,benchmarks\u002Fengines,benchmarks\u002Fds,reports\u002Fcharts,samples\u002Fkitti_output}\n",{"type":43,"tag":84,"props":923,"children":925},{"id":924},"critical-rules",[926],{"type":48,"value":927},"Critical Rules",{"type":43,"tag":929,"props":930,"children":931},"ol",{},[932,958,976,1016,1049,1059,1076,1092,1118,1135,1229],{"type":43,"tag":933,"props":934,"children":935},"li",{},[936,941,943,949,951,957],{"type":43,"tag":57,"props":937,"children":938},{},[939],{"type":48,"value":940},"Engine naming",{"type":48,"value":942}," — always ",{"type":43,"tag":75,"props":944,"children":946},{"className":945},[],[947],{"type":48,"value":948},"{model}_dynamic_b{MAX_BS}.engine",{"type":48,"value":950},". Never bare ",{"type":43,"tag":75,"props":952,"children":954},{"className":953},[],[955],{"type":48,"value":956},"model_dynamic.engine",{"type":48,"value":82},{"type":43,"tag":933,"props":959,"children":960},{},[961,966,968,974],{"type":43,"tag":57,"props":962,"children":963},{},[964],{"type":48,"value":965},"batch_size == num_streams",{"type":48,"value":967}," — in DS runs, ",{"type":43,"tag":75,"props":969,"children":971},{"className":970},[],[972],{"type":48,"value":973},"batch-size",{"type":48,"value":975}," and stream count are always equal.",{"type":43,"tag":933,"props":977,"children":978},{},[979,984,986,992,994,1000,1001,1007,1008,1014],{"type":43,"tag":57,"props":980,"children":981},{},[982],{"type":48,"value":983},"Log filenames are fixed",{"type":48,"value":985}," — ",{"type":43,"tag":75,"props":987,"children":989},{"className":988},[],[990],{"type":48,"value":991},"trtexec_b1.log",{"type":48,"value":993},", ",{"type":43,"tag":75,"props":995,"children":997},{"className":996},[],[998],{"type":48,"value":999},"trtexec_b${MAX_BS}.log",{"type":48,"value":993},{"type":43,"tag":75,"props":1002,"children":1004},{"className":1003},[],[1005],{"type":48,"value":1006},"ds_s${N}_run1.log",{"type":48,"value":993},{"type":43,"tag":75,"props":1009,"children":1011},{"className":1010},[],[1012],{"type":48,"value":1013},"ds_s${N}_run2.log",{"type":48,"value":1015},". No timestamps. Report generation reads exact paths.",{"type":43,"tag":933,"props":1017,"children":1018},{},[1019,1024,1025,1031,1033,1039,1041,1047],{"type":43,"tag":57,"props":1020,"children":1021},{},[1022],{"type":48,"value":1023},"Parser zero-init",{"type":48,"value":942},{"type":43,"tag":75,"props":1026,"children":1028},{"className":1027},[],[1029],{"type":48,"value":1030},"NvDsInferObjectDetectionInfo obj = {};",{"type":48,"value":1032},". Required for DeepStream OBB support; bare ",{"type":43,"tag":75,"props":1034,"children":1036},{"className":1035},[],[1037],{"type":48,"value":1038},"obj;",{"type":48,"value":1040}," leaves ",{"type":43,"tag":75,"props":1042,"children":1044},{"className":1043},[],[1045],{"type":48,"value":1046},"rotation_angle",{"type":48,"value":1048}," uninitialized, causing tilted bounding boxes.",{"type":43,"tag":933,"props":1050,"children":1051},{},[1052,1057],{"type":43,"tag":57,"props":1053,"children":1054},{},[1055],{"type":48,"value":1056},"KITTI validation gate",{"type":48,"value":1058}," — do NOT proceed to Step 7 if KITTI frame count is zero or detection rate \u003C 90%.",{"type":43,"tag":933,"props":1060,"children":1061},{},[1062,1067,1068,1074],{"type":43,"tag":57,"props":1063,"children":1064},{},[1065],{"type":48,"value":1066},"Shared venv",{"type":48,"value":985},{"type":43,"tag":75,"props":1069,"children":1071},{"className":1070},[],[1072],{"type":48,"value":1073},"build\u002F.venv_optimum",{"type":48,"value":1075}," reused across all models. Never create per-model venvs.",{"type":43,"tag":933,"props":1077,"children":1078},{},[1079,1090],{"type":43,"tag":57,"props":1080,"children":1081},{},[1082,1084],{"type":48,"value":1083},"trtexec ",{"type":43,"tag":75,"props":1085,"children":1087},{"className":1086},[],[1088],{"type":48,"value":1089},"--noDataTransfers",{"type":48,"value":1091}," — GPU-only compute matches DeepStream's GPU-to-GPU data flow.",{"type":43,"tag":933,"props":1093,"children":1094},{},[1095,1100,1102,1108,1110,1116],{"type":43,"tag":57,"props":1096,"children":1097},{},[1098],{"type":48,"value":1099},"Report HTML+PDF",{"type":48,"value":1101}," — always use ",{"type":43,"tag":75,"props":1103,"children":1105},{"className":1104},[],[1106],{"type":48,"value":1107},"skills\u002Fdeepstream-import-vision-model\u002Fscripts\u002Freport\u002Fmd-to-html-pdf.py",{"type":48,"value":1109},". Never write a custom HTML generator or call ",{"type":43,"tag":75,"props":1111,"children":1113},{"className":1112},[],[1114],{"type":48,"value":1115},"wkhtmltopdf",{"type":48,"value":1117}," directly.",{"type":43,"tag":933,"props":1119,"children":1120},{},[1121,1126,1128,1133],{"type":43,"tag":57,"props":1122,"children":1123},{},[1124],{"type":48,"value":1125},"Object detection only",{"type":48,"value":1127}," — reject non-detection architectures from ",{"type":43,"tag":75,"props":1129,"children":1131},{"className":1130},[],[1132],{"type":48,"value":80},{"type":48,"value":1134}," before building anything.",{"type":43,"tag":933,"props":1136,"children":1137},{},[1138,1143,1144,1150,1152,1158,1160,1165,1167,1173,1175,1181,1183,1189,1191,1197,1199,1205,1207,1213,1215,1221,1222,1228],{"type":43,"tag":57,"props":1139,"children":1140},{},[1141],{"type":48,"value":1142},"Encoder fallback (MANDATORY)",{"type":48,"value":985},{"type":43,"tag":75,"props":1145,"children":1147},{"className":1146},[],[1148],{"type":48,"value":1149},"x264enc",{"type":48,"value":1151}," and ",{"type":43,"tag":75,"props":1153,"children":1155},{"className":1154},[],[1156],{"type":48,"value":1157},"openh264enc",{"type":48,"value":1159}," are ",{"type":43,"tag":57,"props":1161,"children":1162},{},[1163],{"type":48,"value":1164},"prohibited",{"type":48,"value":1166},". On NVENC-unavailable systems, use ",{"type":43,"tag":75,"props":1168,"children":1170},{"className":1169},[],[1171],{"type":48,"value":1172},"theoraenc + oggmux",{"type":48,"value":1174}," (LGPL; ships in gst-plugins-base; output is ",{"type":43,"tag":75,"props":1176,"children":1178},{"className":1177},[],[1179],{"type":48,"value":1180},".ogv",{"type":48,"value":1182},"). If ",{"type":43,"tag":75,"props":1184,"children":1186},{"className":1185},[],[1187],{"type":48,"value":1188},"theoraenc",{"type":48,"value":1190},"\u002F",{"type":43,"tag":75,"props":1192,"children":1194},{"className":1193},[],[1195],{"type":48,"value":1196},"oggmux",{"type":48,"value":1198}," are absent, skip video creation (",{"type":43,"tag":75,"props":1200,"children":1202},{"className":1201},[],[1203],{"type":48,"value":1204},"DS_SINGLE_STREAM_MODE=skipped",{"type":48,"value":1206},"). Report which mode was used: ",{"type":43,"tag":75,"props":1208,"children":1210},{"className":1209},[],[1211],{"type":48,"value":1212},"nvv4l2h264enc",{"type":48,"value":1214}," \u002F ",{"type":43,"tag":75,"props":1216,"children":1218},{"className":1217},[],[1219],{"type":48,"value":1220},"theoraenc-fallback",{"type":48,"value":1214},{"type":43,"tag":75,"props":1223,"children":1225},{"className":1224},[],[1226],{"type":48,"value":1227},"skipped",{"type":48,"value":82},{"type":43,"tag":933,"props":1230,"children":1231},{},[1232,1237,1239,1245,1247,1253,1255,1261],{"type":43,"tag":57,"props":1233,"children":1234},{},[1235],{"type":48,"value":1236},"Video source (MANDATORY)",{"type":48,"value":1238}," — default is always ",{"type":43,"tag":75,"props":1240,"children":1242},{"className":1241},[],[1243],{"type":48,"value":1244},"sample_720p.mp4",{"type":48,"value":1246}," (1280×720). Never autonomously substitute ",{"type":43,"tag":75,"props":1248,"children":1250},{"className":1249},[],[1251],{"type":48,"value":1252},"sample_1080p_h264.mp4",{"type":48,"value":1254}," or any other file. Only use a different video when the user explicitly provides a path (via ",{"type":43,"tag":75,"props":1256,"children":1258},{"className":1257},[],[1259],{"type":48,"value":1260},"DS_VIDEO",{"type":48,"value":1262}," env var or script argument).",{"type":43,"tag":84,"props":1264,"children":1266},{"id":1265},"pipeline-timing",[1267],{"type":48,"value":1268},"Pipeline Timing",{"type":43,"tag":51,"props":1270,"children":1271},{},[1272],{"type":48,"value":1273},"Wrap every step:",{"type":43,"tag":250,"props":1275,"children":1277},{"className":252,"code":1276,"language":254,"meta":255,"style":255},"STEP_START=$(date +%s.%N)\n# ... step commands ...\nSTEP_END=$(date +%s.%N)\nSTEP_DURATION=$(echo \"$STEP_END - $STEP_START\" | bc)\necho \"[Step N] completed in ${STEP_DURATION}s\"\n",[1278],{"type":43,"tag":75,"props":1279,"children":1280},{"__ignoreMap":255},[1281,1309,1317,1341,1394],{"type":43,"tag":261,"props":1282,"children":1283},{"class":263,"line":264},[1284,1289,1294,1299,1304],{"type":43,"tag":261,"props":1285,"children":1286},{"style":402},[1287],{"type":48,"value":1288},"STEP_START",{"type":43,"tag":261,"props":1290,"children":1291},{"style":312},[1292],{"type":48,"value":1293},"=$(",{"type":43,"tag":261,"props":1295,"children":1296},{"style":278},[1297],{"type":48,"value":1298},"date",{"type":43,"tag":261,"props":1300,"children":1301},{"style":328},[1302],{"type":48,"value":1303}," +%s.%N",{"type":43,"tag":261,"props":1305,"children":1306},{"style":312},[1307],{"type":48,"value":1308},")\n",{"type":43,"tag":261,"props":1310,"children":1311},{"class":263,"line":274},[1312],{"type":43,"tag":261,"props":1313,"children":1314},{"style":268},[1315],{"type":48,"value":1316},"# ... step commands ...\n",{"type":43,"tag":261,"props":1318,"children":1319},{"class":263,"line":284},[1320,1325,1329,1333,1337],{"type":43,"tag":261,"props":1321,"children":1322},{"style":402},[1323],{"type":48,"value":1324},"STEP_END",{"type":43,"tag":261,"props":1326,"children":1327},{"style":312},[1328],{"type":48,"value":1293},{"type":43,"tag":261,"props":1330,"children":1331},{"style":278},[1332],{"type":48,"value":1298},{"type":43,"tag":261,"props":1334,"children":1335},{"style":328},[1336],{"type":48,"value":1303},{"type":43,"tag":261,"props":1338,"children":1339},{"style":312},[1340],{"type":48,"value":1308},{"type":43,"tag":261,"props":1342,"children":1343},{"class":263,"line":294},[1344,1349,1353,1358,1362,1367,1372,1377,1381,1385,1390],{"type":43,"tag":261,"props":1345,"children":1346},{"style":402},[1347],{"type":48,"value":1348},"STEP_DURATION",{"type":43,"tag":261,"props":1350,"children":1351},{"style":312},[1352],{"type":48,"value":1293},{"type":43,"tag":261,"props":1354,"children":1355},{"style":657},[1356],{"type":48,"value":1357},"echo",{"type":43,"tag":261,"props":1359,"children":1360},{"style":312},[1361],{"type":48,"value":445},{"type":43,"tag":261,"props":1363,"children":1364},{"style":402},[1365],{"type":48,"value":1366},"$STEP_END",{"type":43,"tag":261,"props":1368,"children":1369},{"style":328},[1370],{"type":48,"value":1371}," - ",{"type":43,"tag":261,"props":1373,"children":1374},{"style":402},[1375],{"type":48,"value":1376},"$STEP_START",{"type":43,"tag":261,"props":1378,"children":1379},{"style":312},[1380],{"type":48,"value":460},{"type":43,"tag":261,"props":1382,"children":1383},{"style":312},[1384],{"type":48,"value":320},{"type":43,"tag":261,"props":1386,"children":1387},{"style":278},[1388],{"type":48,"value":1389}," bc",{"type":43,"tag":261,"props":1391,"children":1392},{"style":312},[1393],{"type":48,"value":1308},{"type":43,"tag":261,"props":1395,"children":1396},{"class":263,"line":303},[1397,1401,1405,1410,1415,1419,1424,1429],{"type":43,"tag":261,"props":1398,"children":1399},{"style":657},[1400],{"type":48,"value":1357},{"type":43,"tag":261,"props":1402,"children":1403},{"style":312},[1404],{"type":48,"value":445},{"type":43,"tag":261,"props":1406,"children":1407},{"style":328},[1408],{"type":48,"value":1409},"[Step N] completed in ",{"type":43,"tag":261,"props":1411,"children":1412},{"style":312},[1413],{"type":48,"value":1414},"${",{"type":43,"tag":261,"props":1416,"children":1417},{"style":402},[1418],{"type":48,"value":1348},{"type":43,"tag":261,"props":1420,"children":1421},{"style":312},[1422],{"type":48,"value":1423},"}",{"type":43,"tag":261,"props":1425,"children":1426},{"style":328},[1427],{"type":48,"value":1428},"s",{"type":43,"tag":261,"props":1430,"children":1431},{"style":312},[1432],{"type":48,"value":502},{"type":43,"tag":51,"props":1434,"children":1435},{},[1436,1438,1444,1446,1452],{"type":48,"value":1437},"Track ",{"type":43,"tag":75,"props":1439,"children":1441},{"className":1440},[],[1442],{"type":48,"value":1443},"PIPELINE_START",{"type":48,"value":1445}," (before Step 1) and ",{"type":43,"tag":75,"props":1447,"children":1449},{"className":1448},[],[1450],{"type":48,"value":1451},"PIPELINE_END",{"type":48,"value":1453}," (after Step 8). Report all durations in the benchmark report.",{"type":43,"tag":84,"props":1455,"children":1457},{"id":1456},"report-output-mandatory-all-3-formats",[1458],{"type":48,"value":1459},"Report Output (MANDATORY — all 3 formats)",{"type":43,"tag":929,"props":1461,"children":1462},{},[1463,1474,1485],{"type":43,"tag":933,"props":1464,"children":1465},{},[1466,1472],{"type":43,"tag":75,"props":1467,"children":1469},{"className":1468},[],[1470],{"type":48,"value":1471},"benchmark_report.md",{"type":48,"value":1473}," — markdown source (12 mandatory sections)",{"type":43,"tag":933,"props":1475,"children":1476},{},[1477,1483],{"type":43,"tag":75,"props":1478,"children":1480},{"className":1479},[],[1481],{"type":48,"value":1482},"benchmark_report.html",{"type":48,"value":1484}," — styled HTML (charts base64-inlined, no local file access)",{"type":43,"tag":933,"props":1486,"children":1487},{},[1488,1494,1496,1502,1504,1510,1512,1518],{"type":43,"tag":75,"props":1489,"children":1491},{"className":1490},[],[1492],{"type":48,"value":1493},"benchmark_report_{model_name}.pdf",{"type":48,"value":1495}," — via ",{"type":43,"tag":75,"props":1497,"children":1499},{"className":1498},[],[1500],{"type":48,"value":1501},"md-to-html-pdf.py",{"type":48,"value":1503},"; verify charts are embedded by counting ",{"type":43,"tag":75,"props":1505,"children":1507},{"className":1506},[],[1508],{"type":48,"value":1509},"data:image\u002Fpng",{"type":48,"value":1511}," occurrences in the HTML output: ",{"type":43,"tag":75,"props":1513,"children":1515},{"className":1514},[],[1516],{"type":48,"value":1517},"grep -o 'data:image\u002Fpng' benchmark_report.html | wc -l",{"type":48,"value":1519}," should equal 5",{"type":43,"tag":51,"props":1521,"children":1522},{},[1523,1525,1531],{"type":48,"value":1524},"Run charts and report scripts with the shared venv active: ",{"type":43,"tag":75,"props":1526,"children":1528},{"className":1527},[],[1529],{"type":48,"value":1530},"source build\u002F.venv_optimum\u002Fbin\u002Factivate",{"type":48,"value":82},{"type":43,"tag":84,"props":1533,"children":1535},{"id":1534},"reference-documents",[1536],{"type":48,"value":1537},"Reference Documents",{"type":43,"tag":51,"props":1539,"children":1540},{},[1541,1546],{"type":43,"tag":57,"props":1542,"children":1543},{},[1544],{"type":48,"value":1545},"IMPORTANT",{"type":48,"value":1547},": Read the relevant reference before starting each phase. Do NOT generate code from memory.",{"type":43,"tag":91,"props":1549,"children":1550},{},[1551,1567],{"type":43,"tag":95,"props":1552,"children":1553},{},[1554],{"type":43,"tag":99,"props":1555,"children":1556},{},[1557,1562],{"type":43,"tag":103,"props":1558,"children":1559},{},[1560],{"type":48,"value":1561},"Document",{"type":43,"tag":103,"props":1563,"children":1564},{},[1565],{"type":48,"value":1566},"Use When",{"type":43,"tag":124,"props":1568,"children":1569},{},[1570,1585,1600,1615],{"type":43,"tag":99,"props":1571,"children":1572},{},[1573,1580],{"type":43,"tag":131,"props":1574,"children":1575},{},[1576],{"type":43,"tag":145,"props":1577,"children":1578},{"href":147},[1579],{"type":48,"value":147},{"type":43,"tag":131,"props":1581,"children":1582},{},[1583],{"type":48,"value":1584},"Steps 1–3: HF\u002FNGC URL parsing, format detection, ONNX download, SafeTensors export, label extraction",{"type":43,"tag":99,"props":1586,"children":1587},{},[1588,1595],{"type":43,"tag":131,"props":1589,"children":1590},{},[1591],{"type":43,"tag":145,"props":1592,"children":1593},{"href":173},[1594],{"type":48,"value":173},{"type":43,"tag":131,"props":1596,"children":1597},{},[1598],{"type":48,"value":1599},"Steps 4–5: trtexec engine build, benchmarks, PEAK_GPU_STREAMS derivation, iterative scaling",{"type":43,"tag":99,"props":1601,"children":1602},{},[1603,1610],{"type":43,"tag":131,"props":1604,"children":1605},{},[1606],{"type":43,"tag":145,"props":1607,"children":1608},{"href":199},[1609],{"type":48,"value":199},{"type":43,"tag":131,"props":1611,"children":1612},{},[1613],{"type":48,"value":1614},"Steps 6–7: custom bbox parser, nvinfer config, single-stream validation, KITTI dump, multi-stream benchmark",{"type":43,"tag":99,"props":1616,"children":1617},{},[1618,1625],{"type":43,"tag":131,"props":1619,"children":1620},{},[1621],{"type":43,"tag":145,"props":1622,"children":1623},{"href":225},[1624],{"type":48,"value":225},{"type":43,"tag":131,"props":1626,"children":1627},{},[1628],{"type":48,"value":1629},"Step 8: benchmark_data.json, 5 charts, 12-section markdown report, HTML + PDF",{"type":43,"tag":84,"props":1631,"children":1633},{"id":1632},"scripts",[1634],{"type":48,"value":1635},"Scripts",{"type":43,"tag":51,"props":1637,"children":1638},{},[1639,1641,1647],{"type":48,"value":1640},"Located in ",{"type":43,"tag":75,"props":1642,"children":1644},{"className":1643},[],[1645],{"type":48,"value":1646},"scripts\u002F",{"type":48,"value":82},{"type":43,"tag":91,"props":1649,"children":1650},{},[1651,1671],{"type":43,"tag":95,"props":1652,"children":1653},{},[1654],{"type":43,"tag":99,"props":1655,"children":1656},{},[1657,1662,1666],{"type":43,"tag":103,"props":1658,"children":1659},{},[1660],{"type":48,"value":1661},"Script",{"type":43,"tag":103,"props":1663,"children":1664},{},[1665],{"type":48,"value":112},{"type":43,"tag":103,"props":1667,"children":1668},{},[1669],{"type":48,"value":1670},"Purpose",{"type":43,"tag":124,"props":1672,"children":1673},{},[1674,1695,1716,1737,1758,1779,1801,1822,1844,1865,1886,1907,1928,1949,1987,2023,2044,2065,2086,2107,2128,2149],{"type":43,"tag":99,"props":1675,"children":1676},{},[1677,1686,1690],{"type":43,"tag":131,"props":1678,"children":1679},{},[1680],{"type":43,"tag":75,"props":1681,"children":1683},{"className":1682},[],[1684],{"type":48,"value":1685},"model\u002Fhf-list-files.sh",{"type":43,"tag":131,"props":1687,"children":1688},{},[1689],{"type":48,"value":135},{"type":43,"tag":131,"props":1691,"children":1692},{},[1693],{"type":48,"value":1694},"List HuggingFace repo files",{"type":43,"tag":99,"props":1696,"children":1697},{},[1698,1707,1711],{"type":43,"tag":131,"props":1699,"children":1700},{},[1701],{"type":43,"tag":75,"props":1702,"children":1704},{"className":1703},[],[1705],{"type":48,"value":1706},"model\u002Fhf-download-config.sh",{"type":43,"tag":131,"props":1708,"children":1709},{},[1710],{"type":48,"value":135},{"type":43,"tag":131,"props":1712,"children":1713},{},[1714],{"type":48,"value":1715},"Download config.json from HF",{"type":43,"tag":99,"props":1717,"children":1718},{},[1719,1728,1732],{"type":43,"tag":131,"props":1720,"children":1721},{},[1722],{"type":43,"tag":75,"props":1723,"children":1725},{"className":1724},[],[1726],{"type":48,"value":1727},"model\u002Fngc-list-files.sh",{"type":43,"tag":131,"props":1729,"children":1730},{},[1731],{"type":48,"value":135},{"type":43,"tag":131,"props":1733,"children":1734},{},[1735],{"type":48,"value":1736},"List NGC model files",{"type":43,"tag":99,"props":1738,"children":1739},{},[1740,1749,1753],{"type":43,"tag":131,"props":1741,"children":1742},{},[1743],{"type":43,"tag":75,"props":1744,"children":1746},{"className":1745},[],[1747],{"type":48,"value":1748},"model\u002Fngc-download.sh",{"type":43,"tag":131,"props":1750,"children":1751},{},[1752],{"type":48,"value":135},{"type":43,"tag":131,"props":1754,"children":1755},{},[1756],{"type":48,"value":1757},"Download NGC model archive",{"type":43,"tag":99,"props":1759,"children":1760},{},[1761,1770,1774],{"type":43,"tag":131,"props":1762,"children":1763},{},[1764],{"type":43,"tag":75,"props":1765,"children":1767},{"className":1766},[],[1768],{"type":48,"value":1769},"model\u002Fsafetensors-to-onnx.sh",{"type":43,"tag":131,"props":1771,"children":1772},{},[1773],{"type":48,"value":135},{"type":43,"tag":131,"props":1775,"children":1776},{},[1777],{"type":48,"value":1778},"Export SafeTensors → ONNX via optimum-cli",{"type":43,"tag":99,"props":1780,"children":1781},{},[1782,1791,1796],{"type":43,"tag":131,"props":1783,"children":1784},{},[1785],{"type":43,"tag":75,"props":1786,"children":1788},{"className":1787},[],[1789],{"type":48,"value":1790},"model\u002Finspect-onnx.py",{"type":43,"tag":131,"props":1792,"children":1793},{},[1794],{"type":48,"value":1795},"1–5",{"type":43,"tag":131,"props":1797,"children":1798},{},[1799],{"type":48,"value":1800},"Inspect ONNX input\u002Foutput shapes",{"type":43,"tag":99,"props":1802,"children":1803},{},[1804,1813,1817],{"type":43,"tag":131,"props":1805,"children":1806},{},[1807],{"type":43,"tag":75,"props":1808,"children":1810},{"className":1809},[],[1811],{"type":48,"value":1812},"model\u002Fmake-static-batch-onnx.py",{"type":43,"tag":131,"props":1814,"children":1815},{},[1816],{"type":48,"value":162},{"type":43,"tag":131,"props":1818,"children":1819},{},[1820],{"type":48,"value":1821},"Bake batch dim into ONNX",{"type":43,"tag":99,"props":1823,"children":1824},{},[1825,1834,1839],{"type":43,"tag":131,"props":1826,"children":1827},{},[1828],{"type":43,"tag":75,"props":1829,"children":1831},{"className":1830},[],[1832],{"type":48,"value":1833},"model\u002Fcleanup.sh",{"type":43,"tag":131,"props":1835,"children":1836},{},[1837],{"type":48,"value":1838},"Any",{"type":43,"tag":131,"props":1840,"children":1841},{},[1842],{"type":48,"value":1843},"Remove staging dirs, preserve shared venv",{"type":43,"tag":99,"props":1845,"children":1846},{},[1847,1856,1860],{"type":43,"tag":131,"props":1848,"children":1849},{},[1850],{"type":43,"tag":75,"props":1851,"children":1853},{"className":1852},[],[1854],{"type":48,"value":1855},"engine\u002Fbenchmark-trtexec.sh",{"type":43,"tag":131,"props":1857,"children":1858},{},[1859],{"type":48,"value":162},{"type":43,"tag":131,"props":1861,"children":1862},{},[1863],{"type":48,"value":1864},"Run trtexec with standard flags",{"type":43,"tag":99,"props":1866,"children":1867},{},[1868,1877,1881],{"type":43,"tag":131,"props":1869,"children":1870},{},[1871],{"type":43,"tag":75,"props":1872,"children":1874},{"className":1873},[],[1875],{"type":48,"value":1876},"deepstream\u002Fds-single-stream.sh",{"type":43,"tag":131,"props":1878,"children":1879},{},[1880],{"type":48,"value":188},{"type":43,"tag":131,"props":1882,"children":1883},{},[1884],{"type":48,"value":1885},"Single-stream visual validation (NVENC primary; theoraenc+oggmux fallback; skip if neither)",{"type":43,"tag":99,"props":1887,"children":1888},{},[1889,1898,1902],{"type":43,"tag":131,"props":1890,"children":1891},{},[1892],{"type":43,"tag":75,"props":1893,"children":1895},{"className":1894},[],[1896],{"type":48,"value":1897},"deepstream\u002Fds-sweep.sh",{"type":43,"tag":131,"props":1899,"children":1900},{},[1901],{"type":48,"value":188},{"type":43,"tag":131,"props":1903,"children":1904},{},[1905],{"type":48,"value":1906},"2-phase batch size sweep",{"type":43,"tag":99,"props":1908,"children":1909},{},[1910,1919,1923],{"type":43,"tag":131,"props":1911,"children":1912},{},[1913],{"type":43,"tag":75,"props":1914,"children":1916},{"className":1915},[],[1917],{"type":48,"value":1918},"deepstream\u002Fbenchmark-ds.sh",{"type":43,"tag":131,"props":1920,"children":1921},{},[1922],{"type":48,"value":188},{"type":43,"tag":131,"props":1924,"children":1925},{},[1926],{"type":48,"value":1927},"Fixed-stream DS benchmark",{"type":43,"tag":99,"props":1929,"children":1930},{},[1931,1940,1944],{"type":43,"tag":131,"props":1932,"children":1933},{},[1934],{"type":43,"tag":75,"props":1935,"children":1937},{"className":1936},[],[1938],{"type":48,"value":1939},"deepstream\u002Fds-kitti-dump.sh",{"type":43,"tag":131,"props":1941,"children":1942},{},[1943],{"type":48,"value":188},{"type":43,"tag":131,"props":1945,"children":1946},{},[1947],{"type":48,"value":1948},"KITTI detection dump via deepstream-app",{"type":43,"tag":99,"props":1950,"children":1951},{},[1952,1961,1966],{"type":43,"tag":131,"props":1953,"children":1954},{},[1955],{"type":43,"tag":75,"props":1956,"children":1958},{"className":1957},[],[1959],{"type":48,"value":1960},"deepstream\u002Fds-perf-run.sh",{"type":43,"tag":131,"props":1962,"children":1963},{},[1964],{"type":48,"value":1965},"7",{"type":43,"tag":131,"props":1967,"children":1968},{},[1969,1971,1977,1979,1985],{"type":48,"value":1970},"Step 7c two-run benchmark — wraps ",{"type":43,"tag":75,"props":1972,"children":1974},{"className":1973},[],[1975],{"type":48,"value":1976},"deepstream-app",{"type":48,"value":1978}," with ",{"type":43,"tag":75,"props":1980,"children":1982},{"className":1981},[],[1983],{"type":48,"value":1984},"enable-perf-measurement=1",{"type":48,"value":1986},", writes fixed-name log for the report parser",{"type":43,"tag":99,"props":1988,"children":1989},{},[1990,1999,2003],{"type":43,"tag":131,"props":1991,"children":1992},{},[1993],{"type":43,"tag":75,"props":1994,"children":1996},{"className":1995},[],[1997],{"type":48,"value":1998},"deepstream\u002Fextract-frame.sh",{"type":43,"tag":131,"props":2000,"children":2001},{},[2002],{"type":48,"value":188},{"type":43,"tag":131,"props":2004,"children":2005},{},[2006,2008,2014,2016,2021],{"type":48,"value":2007},"Extract sample frames from output video (",{"type":43,"tag":75,"props":2009,"children":2011},{"className":2010},[],[2012],{"type":48,"value":2013},".mp4",{"type":48,"value":2015}," NVENC path or ",{"type":43,"tag":75,"props":2017,"children":2019},{"className":2018},[],[2020],{"type":48,"value":1180},{"type":48,"value":2022}," theoraenc fallback)",{"type":43,"tag":99,"props":2024,"children":2025},{},[2026,2035,2039],{"type":43,"tag":131,"props":2027,"children":2028},{},[2029],{"type":43,"tag":75,"props":2030,"children":2032},{"className":2031},[],[2033],{"type":48,"value":2034},"report\u002Fgenerate-benchmark-charts.py",{"type":43,"tag":131,"props":2036,"children":2037},{},[2038],{"type":48,"value":214},{"type":43,"tag":131,"props":2040,"children":2041},{},[2042],{"type":48,"value":2043},"Generate 5 benchmark PNG charts",{"type":43,"tag":99,"props":2045,"children":2046},{},[2047,2056,2060],{"type":43,"tag":131,"props":2048,"children":2049},{},[2050],{"type":43,"tag":75,"props":2051,"children":2053},{"className":2052},[],[2054],{"type":48,"value":2055},"report\u002Fmd-to-html-pdf.py",{"type":43,"tag":131,"props":2057,"children":2058},{},[2059],{"type":48,"value":214},{"type":43,"tag":131,"props":2061,"children":2062},{},[2063],{"type":48,"value":2064},"Markdown → styled HTML → PDF (canonical benchmark report path)",{"type":43,"tag":99,"props":2066,"children":2067},{},[2068,2077,2081],{"type":43,"tag":131,"props":2069,"children":2070},{},[2071],{"type":43,"tag":75,"props":2072,"children":2074},{"className":2073},[],[2075],{"type":48,"value":2076},"report\u002Fmd-to-pdf.sh",{"type":43,"tag":131,"props":2078,"children":2079},{},[2080],{"type":48,"value":1838},{"type":43,"tag":131,"props":2082,"children":2083},{},[2084],{"type":48,"value":2085},"Markdown → PDF via pandoc\u002Fpdflatex — for design docs and references only, NOT for benchmark reports (use md-to-html-pdf.py for those)",{"type":43,"tag":99,"props":2087,"children":2088},{},[2089,2098,2102],{"type":43,"tag":131,"props":2090,"children":2091},{},[2092],{"type":43,"tag":75,"props":2093,"children":2095},{"className":2094},[],[2096],{"type":48,"value":2097},"report\u002Freport-style.css",{"type":43,"tag":131,"props":2099,"children":2100},{},[2101],{"type":48,"value":214},{"type":43,"tag":131,"props":2103,"children":2104},{},[2105],{"type":48,"value":2106},"CSS for HTML report",{"type":43,"tag":99,"props":2108,"children":2109},{},[2110,2119,2123],{"type":43,"tag":131,"props":2111,"children":2112},{},[2113],{"type":43,"tag":75,"props":2114,"children":2116},{"className":2115},[],[2117],{"type":48,"value":2118},"report\u002Frender-mermaid-for-pdf.py",{"type":43,"tag":131,"props":2120,"children":2121},{},[2122],{"type":48,"value":214},{"type":43,"tag":131,"props":2124,"children":2125},{},[2126],{"type":48,"value":2127},"Mermaid diagram → PNG",{"type":43,"tag":99,"props":2129,"children":2130},{},[2131,2140,2144],{"type":43,"tag":131,"props":2132,"children":2133},{},[2134],{"type":43,"tag":75,"props":2135,"children":2137},{"className":2136},[],[2138],{"type":48,"value":2139},"report\u002Fmermaid-puppeteer.json",{"type":43,"tag":131,"props":2141,"children":2142},{},[2143],{"type":48,"value":214},{"type":43,"tag":131,"props":2145,"children":2146},{},[2147],{"type":48,"value":2148},"Vetted Puppeteer config for Mermaid (sandboxed; non-root)",{"type":43,"tag":99,"props":2150,"children":2151},{},[2152,2161,2165],{"type":43,"tag":131,"props":2153,"children":2154},{},[2155],{"type":43,"tag":75,"props":2156,"children":2158},{"className":2157},[],[2159],{"type":48,"value":2160},"report\u002Fmermaid-puppeteer-root.json",{"type":43,"tag":131,"props":2162,"children":2163},{},[2164],{"type":48,"value":214},{"type":43,"tag":131,"props":2166,"children":2167},{},[2168],{"type":48,"value":2169},"Vetted Puppeteer config for Mermaid (used when running as root)",{"type":43,"tag":84,"props":2171,"children":2173},{"id":2172},"quick-error-reference",[2174],{"type":48,"value":2175},"Quick Error Reference",{"type":43,"tag":91,"props":2177,"children":2178},{},[2179,2195],{"type":43,"tag":95,"props":2180,"children":2181},{},[2182],{"type":43,"tag":99,"props":2183,"children":2184},{},[2185,2190],{"type":43,"tag":103,"props":2186,"children":2187},{},[2188],{"type":48,"value":2189},"Error",{"type":43,"tag":103,"props":2191,"children":2192},{},[2193],{"type":48,"value":2194},"Fix",{"type":43,"tag":124,"props":2196,"children":2197},{},[2198,2216,2250,2277,2304,2345,2366,2387],{"type":43,"tag":99,"props":2199,"children":2200},{},[2201,2206],{"type":43,"tag":131,"props":2202,"children":2203},{},[2204],{"type":48,"value":2205},"Tilted\u002Fdiagonal bounding boxes",{"type":43,"tag":131,"props":2207,"children":2208},{},[2209,2211],{"type":48,"value":2210},"Parser struct not zero-initialized — use ",{"type":43,"tag":75,"props":2212,"children":2214},{"className":2213},[],[2215],{"type":48,"value":1030},{"type":43,"tag":99,"props":2217,"children":2218},{},[2219,2224],{"type":43,"tag":131,"props":2220,"children":2221},{},[2222],{"type":48,"value":2223},"Zero KITTI files",{"type":43,"tag":131,"props":2225,"children":2226},{},[2227,2233,2235,2241,2243,2248],{"type":43,"tag":75,"props":2228,"children":2230},{"className":2229},[],[2231],{"type":48,"value":2232},"gie-kitti-output-dir",{"type":48,"value":2234}," not read by nvinfer — use ",{"type":43,"tag":75,"props":2236,"children":2238},{"className":2237},[],[2239],{"type":48,"value":2240},"ds-kitti-dump.sh",{"type":48,"value":2242}," (wraps ",{"type":43,"tag":75,"props":2244,"children":2246},{"className":2245},[],[2247],{"type":48,"value":1976},{"type":48,"value":2249},")",{"type":43,"tag":99,"props":2251,"children":2252},{},[2253,2258],{"type":43,"tag":131,"props":2254,"children":2255},{},[2256],{"type":48,"value":2257},"Engine rebuilds every DS run",{"type":43,"tag":131,"props":2259,"children":2260},{},[2261,2267,2269,2275],{"type":43,"tag":75,"props":2262,"children":2264},{"className":2263},[],[2265],{"type":48,"value":2266},"model-engine-file",{"type":48,"value":2268}," path wrong — check relative path from ",{"type":43,"tag":75,"props":2270,"children":2272},{"className":2271},[],[2273],{"type":48,"value":2274},"config\u002F",{"type":48,"value":2276}," dir",{"type":43,"tag":99,"props":2278,"children":2279},{},[2280,2291],{"type":43,"tag":131,"props":2281,"children":2282},{},[2283,2289],{"type":43,"tag":75,"props":2284,"children":2286},{"className":2285},[],[2287],{"type":48,"value":2288},"setDimensions",{"type":48,"value":2290}," negative dims",{"type":43,"tag":131,"props":2292,"children":2293},{},[2294,2296,2302],{"type":48,"value":2295},"Add ",{"type":43,"tag":75,"props":2297,"children":2299},{"className":2298},[],[2300],{"type":48,"value":2301},"infer-dims=3;H;W",{"type":48,"value":2303}," to nvinfer config for dynamic ONNX models",{"type":43,"tag":99,"props":2305,"children":2306},{},[2307,2318],{"type":43,"tag":131,"props":2308,"children":2309},{},[2310,2316],{"type":43,"tag":75,"props":2311,"children":2313},{"className":2312},[],[2314],{"type":48,"value":2315},"--memPoolSize",{"type":48,"value":2317}," workspace 0.03 MiB",{"type":43,"tag":131,"props":2319,"children":2320},{},[2321,2323,2329,2331,2337,2339],{"type":48,"value":2322},"Use ",{"type":43,"tag":75,"props":2324,"children":2326},{"className":2325},[],[2327],{"type":48,"value":2328},"M",{"type":48,"value":2330}," suffix not ",{"type":43,"tag":75,"props":2332,"children":2334},{"className":2333},[],[2335],{"type":48,"value":2336},"MiB",{"type":48,"value":2338}," — e.g. ",{"type":43,"tag":75,"props":2340,"children":2342},{"className":2341},[],[2343],{"type":48,"value":2344},"--memPoolSize=workspace:32768M",{"type":43,"tag":99,"props":2346,"children":2347},{},[2348,2353],{"type":43,"tag":131,"props":2349,"children":2350},{},[2351],{"type":48,"value":2352},"ForeignNode build failure (DETR)",{"type":43,"tag":131,"props":2354,"children":2355},{},[2356,2358,2364],{"type":48,"value":2357},"Use dynamo export path or run ",{"type":43,"tag":75,"props":2359,"children":2361},{"className":2360},[],[2362],{"type":48,"value":2363},"onnxsim",{"type":48,"value":2365}," — see references\u002Fengine-build.md",{"type":43,"tag":99,"props":2367,"children":2368},{},[2369,2374],{"type":43,"tag":131,"props":2370,"children":2371},{},[2372],{"type":48,"value":2373},"Zero detections",{"type":43,"tag":131,"props":2375,"children":2376},{},[2377,2379,2385],{"type":48,"value":2378},"Wrong ",{"type":43,"tag":75,"props":2380,"children":2382},{"className":2381},[],[2383],{"type":48,"value":2384},"net-scale-factor",{"type":48,"value":2386}," — check model family table in references\u002Fpipeline-run.md",{"type":43,"tag":99,"props":2388,"children":2389},{},[2390,2399],{"type":43,"tag":131,"props":2391,"children":2392},{},[2393],{"type":43,"tag":75,"props":2394,"children":2396},{"className":2395},[],[2397],{"type":48,"value":2398},"No module named 'pyservicemaker'",{"type":43,"tag":131,"props":2400,"children":2401},{},[2402,2404],{"type":48,"value":2403},"Install into venv: ",{"type":43,"tag":75,"props":2405,"children":2407},{"className":2406},[],[2408],{"type":48,"value":2409},"pip install \u002Fopt\u002Fnvidia\u002Fdeepstream\u002F...\u002Fpyservicemaker*.whl",{"type":43,"tag":2411,"props":2412,"children":2413},"style",{},[2414],{"type":48,"value":2415},"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":2417,"total":2519},[2418,2435,2449,2463,2475,2490,2505],{"slug":2419,"name":2419,"fn":2420,"description":2421,"org":2422,"tags":2423,"stars":23,"repoUrl":24,"updatedAt":2434},"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},[2424,2427,2430,2431],{"name":2425,"slug":2426,"type":15},"Data Analysis","data-analysis",{"name":2428,"slug":2429,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":2432,"slug":2433,"type":15},"Performance","performance","2026-07-14T05:28:43.176466",{"slug":2436,"name":2436,"fn":2437,"description":2438,"org":2439,"tags":2440,"stars":23,"repoUrl":24,"updatedAt":2448},"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},[2441,2444,2447],{"name":2442,"slug":2443,"type":15},"Deployment","deployment",{"name":2445,"slug":2446,"type":15},"Infrastructure","infrastructure",{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":2450,"name":2450,"fn":2451,"description":2452,"org":2453,"tags":2454,"stars":23,"repoUrl":24,"updatedAt":2462},"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},[2455,2458,2459],{"name":2456,"slug":2457,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":2460,"slug":2461,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":2464,"name":2464,"fn":2465,"description":2466,"org":2467,"tags":2468,"stars":23,"repoUrl":24,"updatedAt":2474},"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},[2469,2470,2471],{"name":2425,"slug":2426,"type":15},{"name":9,"slug":8,"type":15},{"name":2472,"slug":2473,"type":15},"Testing","testing","2026-07-17T05:29:03.913266",{"slug":2476,"name":2476,"fn":2477,"description":2478,"org":2479,"tags":2480,"stars":23,"repoUrl":24,"updatedAt":2489},"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},[2481,2482,2485,2486],{"name":17,"slug":18,"type":15},{"name":2483,"slug":2484,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":2487,"slug":2488,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":2491,"name":2491,"fn":2492,"description":2493,"org":2494,"tags":2495,"stars":23,"repoUrl":24,"updatedAt":2504},"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},[2496,2497,2500,2501],{"name":2442,"slug":2443,"type":15},{"name":2498,"slug":2499,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":2502,"slug":2503,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":2506,"name":2506,"fn":2507,"description":2508,"org":2509,"tags":2510,"stars":23,"repoUrl":24,"updatedAt":2518},"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},[2511,2512,2515],{"name":9,"slug":8,"type":15},{"name":2513,"slug":2514,"type":15},"Quantum Computing","quantum-computing",{"name":2516,"slug":2517,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305,{"items":2521,"total":2672},[2522,2540,2556,2567,2579,2593,2606,2620,2631,2640,2654,2663],{"slug":2523,"name":2523,"fn":2524,"description":2525,"org":2526,"tags":2527,"stars":2537,"repoUrl":2538,"updatedAt":2539},"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},[2528,2531,2534],{"name":2529,"slug":2530,"type":15},"Documentation","documentation",{"name":2532,"slug":2533,"type":15},"MCP","mcp",{"name":2535,"slug":2536,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":2541,"name":2541,"fn":2542,"description":2543,"org":2544,"tags":2545,"stars":2553,"repoUrl":2554,"updatedAt":2555},"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},[2546,2549,2550],{"name":2547,"slug":2548,"type":15},"Containers","containers",{"name":2442,"slug":2443,"type":15},{"name":2551,"slug":2552,"type":15},"Python","python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":2557,"name":2557,"fn":2558,"description":2559,"org":2560,"tags":2561,"stars":2553,"repoUrl":2554,"updatedAt":2566},"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},[2562,2565],{"name":2563,"slug":2564,"type":15},"CI\u002FCD","ci-cd",{"name":2442,"slug":2443,"type":15},"2026-07-14T05:25:59.97109",{"slug":2568,"name":2568,"fn":2569,"description":2570,"org":2571,"tags":2572,"stars":2553,"repoUrl":2554,"updatedAt":2578},"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},[2573,2574,2575],{"name":2563,"slug":2564,"type":15},{"name":2442,"slug":2443,"type":15},{"name":2576,"slug":2577,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":2580,"name":2580,"fn":2581,"description":2582,"org":2583,"tags":2584,"stars":2553,"repoUrl":2554,"updatedAt":2592},"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},[2585,2588,2589],{"name":2586,"slug":2587,"type":15},"Debugging","debugging",{"name":2576,"slug":2577,"type":15},{"name":2590,"slug":2591,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":2594,"name":2594,"fn":2595,"description":2596,"org":2597,"tags":2598,"stars":2553,"repoUrl":2554,"updatedAt":2605},"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},[2599,2602],{"name":2600,"slug":2601,"type":15},"Best Practices","best-practices",{"name":2603,"slug":2604,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":2607,"name":2607,"fn":2608,"description":2609,"org":2610,"tags":2611,"stars":2553,"repoUrl":2554,"updatedAt":2619},"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},[2612,2615,2618],{"name":2613,"slug":2614,"type":15},"Machine Learning","machine-learning",{"name":2616,"slug":2617,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":2621,"name":2621,"fn":2622,"description":2623,"org":2624,"tags":2625,"stars":2553,"repoUrl":2554,"updatedAt":2630},"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},[2626,2629],{"name":2627,"slug":2628,"type":15},"QA","qa",{"name":2472,"slug":2473,"type":15},"2026-07-14T05:25:53.673039",{"slug":2632,"name":2632,"fn":2633,"description":2634,"org":2635,"tags":2636,"stars":2553,"repoUrl":2554,"updatedAt":2639},"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},[2637,2638],{"name":2442,"slug":2443,"type":15},{"name":2445,"slug":2446,"type":15},"2026-07-14T05:25:49.362534",{"slug":2641,"name":2641,"fn":2642,"description":2643,"org":2644,"tags":2645,"stars":2553,"repoUrl":2554,"updatedAt":2653},"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},[2646,2649,2650],{"name":2647,"slug":2648,"type":15},"Code Review","code-review",{"name":2576,"slug":2577,"type":15},{"name":2651,"slug":2652,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":2655,"name":2655,"fn":2656,"description":2657,"org":2658,"tags":2659,"stars":2553,"repoUrl":2554,"updatedAt":2662},"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},[2660,2661],{"name":2627,"slug":2628,"type":15},{"name":2472,"slug":2473,"type":15},"2026-07-14T05:25:54.928983",{"slug":2664,"name":2664,"fn":2665,"description":2666,"org":2667,"tags":2668,"stars":2553,"repoUrl":2554,"updatedAt":2671},"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},[2669,2670],{"name":17,"slug":18,"type":15},{"name":2563,"slug":2564,"type":15},"2026-07-30T05:29:03.275638",496]