[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-doca-flow-dpa-perf":3,"mdc-we7ggv-key":34,"related-org-nvidia-doca-flow-dpa-perf":1380,"related-repo-nvidia-doca-flow-dpa-perf":1541},{"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},"doca-flow-dpa-perf","measure DPA-offloaded DOCA Flow performance","Use this skill when the user is invoking doca_flow_dpa_perf on DPA-capable hardware (ConnectX-7 minimum supported, ConnectX-8 recommended, or BlueField-3) to measure rule update \u002F disable rates on the DPA-offloaded DOCA Flow path — picking the active \u002F passive device split, choosing workload-shape axes (burst, queue, completion threshold, workers, hash pipe algo, PSL tables), or reading Kops\u002Fsec iteration stats and the optional self-test. Trigger even when the user does not explicitly mention \"doca_flow_dpa_perf\" or \"DPA Provider\" — typical implicit phrasings include \"how fast can the DPA program path-selector entries\", \"baseline rule-update rate on ConnectX-8\", \"tool reports zero ops on my BlueField\", \"self-test sentinel never shows on tcpdump\", or \"is my BlueField-2 DPA-capable\". Refuse and route elsewhere for the host \u002F DPU-CPU Flow path (doca-flow-perf), Flow pipeline tuning (doca-flow-tune), writing doca-flow \u002F doca-dpa applications, or DOCA install — those belong to other skills.\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},"Performance","performance","tag",{"name":17,"slug":18,"type":15},"Monitoring","monitoring",{"name":9,"slug":8,"type":15},{"name":21,"slug":22,"type":15},"Engineering","engineering",2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-07-30T05:28:42.498817","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\u002Fdoca-flow-dpa-perf","---\nlicense: Apache-2.0\nname: doca-flow-dpa-perf\ndescription: >\n  Use this skill when the user is invoking doca_flow_dpa_perf on\n  DPA-capable hardware (ConnectX-7 minimum supported,\n  ConnectX-8 recommended, or BlueField-3) to measure rule\n  update \u002F disable rates on the DPA-offloaded DOCA Flow path —\n  picking the active \u002F passive device split, choosing workload-shape\n  axes (burst, queue, completion threshold, workers, hash pipe algo,\n  PSL tables), or reading Kops\u002Fsec iteration stats and the optional\n  self-test. Trigger even when the user does not explicitly mention\n  \"doca_flow_dpa_perf\" or \"DPA Provider\" — typical implicit phrasings\n  include \"how fast can the DPA program path-selector entries\",\n  \"baseline rule-update rate on ConnectX-8\", \"tool reports zero ops\n  on my BlueField\", \"self-test sentinel never shows on tcpdump\", or\n  \"is my BlueField-2 DPA-capable\". Refuse and route elsewhere for the\n  host \u002F DPU-CPU Flow path (doca-flow-perf), Flow pipeline tuning\n  (doca-flow-tune), writing doca-flow \u002F doca-dpa applications, or\n  DOCA install — those belong to other skills.\nmetadata:\n  kind: tool\ncompatibility: >\n  Requires DOCA SDK installed at \u002Fopt\u002Fmellanox\u002Fdoca on Linux (Ubuntu\n  22.04\u002F24.04 or RHEL\u002FSLES) with a DPA-capable device attached —\n  ConnectX-7 as the minimum supported ConnectX generation,\n  ConnectX-8 recommended, or BlueField-3 (BlueField-2 and earlier\n  ConnectX are unsupported). VNF Flow mode required; PF or VF only (SFs are not\n  supported on the DPA path). Reads `pkg-config doca-flow` and the\n  shipped `doca_flow_dpa_perf` binary plus its README on the user's\n  install.\n---\n\n# DOCA Flow DPA Perf (`doca_flow_dpa_perf`)\n\n**Where to start:** This is a tool skill for invoking\n`doca_flow_dpa_perf`, the DPA-accelerated Flow performance tool.\nOpen [`TASKS.md`](TASKS.md) and start at\n[`## configure`](TASKS.md#configure) to confirm DPA-capable\nhardware + VNF Flow mode + the active \u002F passive device split, then\n[`## run`](TASKS.md#run) for the smoke-before-bulk flow with a\nsmall operation count before any sweep, then\n[`## test`](TASKS.md#test) for the eval-loop overlay that gates\ndefensible Kops\u002Fsec numbers. Open [`CAPABILITIES.md`](CAPABILITIES.md)\nwhen the question is *what `doca_flow_dpa_perf` can measure*,\n*what the DPA preconditions are*, *which devices it runs on*,\nor *how to interpret update \u002F disable \u002F self-test output without\nfooling yourself*. If DOCA is not installed yet, route to\n[`doca-setup`](..\u002F..\u002Fdoca-setup\u002FSKILL.md) first; if the device is\nnot DPA-capable (no ConnectX-7+ or BlueField-3+) then this tool is\nthe wrong surface and the right answer is\n[`doca-flow-perf`](..\u002Fdoca-flow-perf\u002FSKILL.md).\n\n## Example questions this skill answers well\n\nThe CLASSES of `doca_flow_dpa_perf` questions this skill is built\nto answer, each with one worked example. The class is the\nload-bearing piece; the worked example is one instance.\n\n- **\"Should I measure the DPA-offloaded Flow path or the\n  host \u002F DPU-CPU Flow path for this question?\"** — worked\n  example: *\"my workload programs path-selector entries via\n  DOCA Flow; do I baseline with `doca_flow_dpa_perf` or with\n  `doca_flow_perf`?\"*. Answered by the *DPA-vs-host* boundary\n  in\n  [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes)\n  and the device-preconditions table.\n- **\"What does the DPA-offload actually accelerate, and what\n  doesn't it change?\"** — worked example: *\"if I move my Flow\n  rule update path to the DPA, what changes in the data plane\n  for the packets themselves?\"*. Answered by the DPA-Provider\n  scope in\n  [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes).\n- **\"What hardware do I need to use this tool at all?\"** —\n  worked example: *\"is my BlueField-2 DPA-capable?\"*. Answered\n  by the device-preconditions table in\n  [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes)\n  (BlueField-3 yes, BlueField-2 no; ConnectX-7 minimum\n  supported, ConnectX-8 recommended, and later generations\n  supported per the public guide and the\n  shipped README on the user's install).\n- **\"How do I size my run — burst, queue, completion threshold,\n  number of operations, iterations — to get a defensible\n  Kops\u002Fsec number?\"** — worked example: *\"I want the median\n  iteration time and standard deviation, not a single noisy\n  first-iteration spike\"*. Answered by the eval-loop overlay\n  in\n  [`TASKS.md ## test`](TASKS.md#test) and the iteration-stats\n  rule in\n  [`CAPABILITIES.md ## Observability`](CAPABILITIES.md#observability).\n- **\"My tool reports zero ops \u002F hangs \u002F fails the self-test —\n  what does that mean?\"** — worked example: *\"the tool runs but\n  the self-test step fails\"*. Answered by the layered error\n  taxonomy in\n  [`CAPABILITIES.md ## Error taxonomy`](CAPABILITIES.md#error-taxonomy)\n  + the debug ladder in\n  [`TASKS.md ## debug`](TASKS.md#debug).\n- **\"How do I quote a DPA-perf number alongside a host-side\n  Flow-perf number for the same workload, in a way the next\n  engineer can actually compare?\"** — worked example: *\"two\n  Kops\u002Fsec numbers for what is supposedly the same workload\"*.\n  Answered by the four-tuple capture rule in\n  [`CAPABILITIES.md ## Safety policy`](CAPABILITIES.md#safety-policy)\n  + the per-tool-name rule (the host tool and the DPA tool are\n  different surfaces; their numbers are not interchangeable\n  without naming which tool produced which).\n\n## Audience\n\nThis skill serves **external operators, performance engineers,\nDOCA Flow application developers, and AI agents who need a\ndefensible measurement of the DPA-offloaded Flow update path** on\nDPA-capable hardware. Concretely:\n\n- A platform operator deciding whether to move a path-selector\n  workload onto the DPA versus keeping it on the host \u002F DPU-CPU\n  path, and wanting a number to compare.\n- A performance engineer producing a *\"DPA Kops\u002Fsec for update\n  operation, queue-size X, burst-size Y, N workers\"* baseline\n  on a specific device + DOCA version so a downstream\n  comparison is meaningful.\n- A DOCA Flow application developer who has already used\n  `doca-dpa` to land a DPA-offload of their Flow rule update\n  path and wants to characterize what the device delivers.\n- An AI agent answering *\"what update rate should I expect from\n  the DPA-offloaded Flow path on device Y?\"* honestly — with a\n  measured number, the command line that produced it, and the\n  device + DOCA version + as-deployed environment that scopes\n  it — instead of guessing from datasheet headlines.\n\nIt is **not** for users debugging the tool's source code,\n**not** a substitute for the live public DOCA Flow DPA Perf guide\non `docs.nvidia.com`, **not** the place to learn the `doca-flow`\nor `doca-dpa` APIs (that audience belongs in\n[`doca-flow`](..\u002F..\u002Flibs\u002Fdoca-flow\u002FSKILL.md) and\n`doca-dpa`), and **not** the right\ntool for the host \u002F DPU-CPU Flow path (route to\n[`doca-flow-perf`](..\u002Fdoca-flow-perf\u002FSKILL.md)).\n\n`doca_flow_dpa_perf` is shipped as a **single CLI binary** with\nDPA-side device code linked in. The skill uses the same\n`kind: tool` three-file shape as the rest of the bundle so\nthe agent's task-verb contract is uniform across the bundle.\n\n## Language scope\n\nThis skill governs invocation, output interpretation, and\nrecommendation-of-routing for the `doca_flow_dpa_perf` CLI on\nDPA-capable hardware. The tool itself has both a host-side\ncontrol (C-language ARGP + DOCA + DPDK code per the shipped\n`flow_dpa_perf.c` \u002F `flow_dpa_perf_core.c`) and a DPA-side device\ncomponent (DPA-side code on the shipped DPA device runtime).\nExternal users do not link any of this; what they configure is\nthe JSON-config-or-CLI invocation surface. For the\n`doca-dpa` programming model behind the DPA-side execution\nengine, see\n`doca-dpa`; for the `doca-flow`\nAPI behind the pipeline the DPA path executes, see\n[`doca-flow`](..\u002F..\u002Flibs\u002Fdoca-flow\u002FSKILL.md).\n\n## When to load this skill\n\nLoad this skill when the user is — or the agent needs to —\ninvoke `doca_flow_dpa_perf` on a real host with DOCA installed\nand a DPA-capable device attached (or the public NGC DOCA\ncontainer with the equivalent device passthrough) to measure\nupdate \u002F disable rates on the DPA-offloaded Flow path.\nConcretely:\n\n- Confirming DPA preconditions (DPA-capable device class,\n  VNF Flow mode, recommended PF use, no SFs) before invoking\n  the tool.\n- Picking the active \u002F passive device split appropriate to the\n  user's hardware (two-port BlueField-3 active + passive; one-\n  port ConnectX-9 active only).\n- Picking the workload-shape axes (burst size, queue size,\n  completion threshold, hash pipe algorithm, work policy,\n  number of PSL tables, table size, number of workers).\n- Picking the operation axis (update or disable-enable) per the\n  shipped README's documented operations.\n- Producing a defensible Kops\u002Fsec number with iteration stats\n  (median, max, standard deviation) captured.\n- Diagnosing zero-ops \u002F hung \u002F failed-self-test runs through\n  the layered error taxonomy.\n\nDo **not** load this skill for general DOCA orientation, Flow\nprogram API work, or installation. For those, use\n[`doca-public-knowledge-map`](..\u002F..\u002Fdoca-public-knowledge-map\u002FSKILL.md),\nthe matching `libs\u002F\u003Clibrary>` skill, or\n[`doca-setup`](..\u002F..\u002Fdoca-setup\u002FSKILL.md). Do not load it for\nthe host \u002F DPU-CPU Flow path — that audience belongs in\n[`doca-flow-perf`](..\u002Fdoca-flow-perf\u002FSKILL.md).\n\n## What this skill provides\n\nThis is a **thin loader**. Substantive material lives in two\ncompanion files:\n\n- `CAPABILITIES.md` — what `doca_flow_dpa_perf` measures\n  (the DPA-Provider-on-DPA-device update \u002F disable path\n  specifically), the DPA-vs-host-path boundary, the\n  device-preconditions table (ConnectX-7+ \u002F BlueField-3+),\n  the documented VNF-only Flow-mode rule, the PF-vs-VF-vs-SF\n  rule (SFs not supported on DPA), the workload-shape axes\n  (burst, queue, completion threshold, hash pipe algorithm,\n  work policy, PSL tables, table size, workers), the\n  operation axis (update vs disable-enable), the version\n  overlay (this tool rides the `doca-flow` and `doca-dpa`\n  versions it links against; the canonical rules live in\n  [`doca-version`](..\u002F..\u002Fdoca-version\u002FSKILL.md)), the layered\n  error taxonomy\n  (config-syntax \u002F device-binding \u002F dpa-precondition \u002F\n  workload-precondition \u002F measurement-soundness \u002F self-test \u002F\n  version \u002F cross-cutting), the observability surface\n  (iteration statistics, self-test path-selector verification,\n  tcpdump-side traffic verification), and the safety posture\n  (smoke-before-bulk, four-tuple capture, name the tool that\n  produced the number).\n- `TASKS.md` — step-by-step workflows for the in-scope task\n  verbs: `install` (route to setup; the binary is shipped),\n  `configure` (DPA-preconditions + active \u002F passive device +\n  workload-shape decision), `build` (route to install — the\n  binary is shipped), `modify` (refuse — modify the invocation,\n  not the binary), `run` (smoke before bulk), `test` (eval\n  loop), `debug` (layered diagnosis), `use` (consume the\n  captured number), plus a `Deferred task verbs` block routing\n  out-of-scope questions and a `Command appendix`.\n\nThe skill assumes a host where DOCA is already installed (or\nthe NGC DOCA container is running) on a DPA-capable device and\nthe operator has the permissions to bind the device and allocate\nthe DPA execution resources the tool needs.\n\n## What this skill deliberately does not ship\n\nThis skill is **agent guidance**, not a samples or scripts\nbundle. To keep the boundary clean, it deliberately does not\ncontain — and pull requests should not add:\n\n- **Verbatim default values for flag inventories beyond what\n  the shipped README or installed `--help` documents.** Read\n  defaults from the README first, then fall back to the\n  installed binary's `--help`. If neither defines a needed\n  default, stop and request the operator's explicit value\n  instead of guessing. The\n  flag surface is install-specific within the documented\n  surface; the documented invocations + `--help` on the\n  installed version are the authoritative answer. Inventing\n  a flag is the most common hallucination failure.\n- **Pre-baked example Kops\u002Fsec numbers or expected throughput\n  numbers.** Output is device-, firmware-, DOCA-version-,\n  workload-, and platform-specific; a pinned number for one\n  platform misleads operators on a different platform \u002F\n  version. The shipped README's example numbers are\n  *illustrative*, not a baseline the agent should quote as\n  ground truth.\n- **Wrappers, parsers, or scripts** in any language that\n  consume the tool's stdout \u002F CSV. The output format is\n  documented; if a user wants to script against it, the\n  right answer is \"read the live guide, write the parser\n  against your installed version\".\n- **A `samples\u002F` or `reference\u002F` subtree.** This is a thin\n  loader for a documented CLI; substantive material lives on\n  the public page, in `--help`, and in the shipped README on\n  the user's install.\n\n## Loading order\n\n1. Read this `SKILL.md` first to confirm the user's question\n   is in scope (the user actually wants to invoke\n   `doca_flow_dpa_perf` on DPA-capable hardware, not measure\n   the host \u002F DPU-CPU Flow path).\n2. **For what `doca_flow_dpa_perf` measures, the DPA-vs-host\n   boundary, the device-preconditions table, the workload-\n   shape axes, the version overlay, the error taxonomy, the\n   observability surface, and the safety posture, see\n   [CAPABILITIES.md](CAPABILITIES.md).**\n3. **For the documented invocations and the smoke-before-bulk\n   workflow — `install`, `configure`, `build`, `modify`,\n   `run`, `test`, `debug`, `use` — see [TASKS.md](TASKS.md).**\n\n## Related skills\n\n- [`doca-flow`](..\u002F..\u002Flibs\u002Fdoca-flow\u002FSKILL.md) — the **base\n  library** whose pipeline this tool measures on the DPA\n  path. The pipe \u002F entry \u002F rule surface this tool drives is\n  created by `doca-flow` program code; the library's pipe\n  attributes and capability surface are the upstream context.\n- `doca-dpa` — the\n  programming model behind the DPA execution engine the tool\n  runs on. When the user's question goes from *\"measure the\n  DPA path\"* to *\"why is the DPA path doing this\"*, that\n  skill is the next stop.\n- [`doca-flow-perf`](..\u002Fdoca-flow-perf\u002FSKILL.md) — the\n  host \u002F DPU-CPU Flow performance tool. The cross-tool\n  comparison rule lives in\n  [`CAPABILITIES.md ## Capabilities and modes`](CAPABILITIES.md#capabilities-and-modes):\n  name which tool produced which number.\n- [`doca-flow-tune`](..\u002Fdoca-flow-tune\u002FSKILL.md) — the Flow\n  tuning tool. A DPA-perf number is the kind of baseline\n  `doca-flow-tune` then optimizes on top of, via a Flow-program\n  modify-a-sample loop.\n- [`doca-public-knowledge-map`](..\u002F..\u002Fdoca-public-knowledge-map\u002FSKILL.md) —\n  routing to the public DOCA Flow DPA Perf page on\n  `docs.nvidia.com` and the rest of the public DOCA\n  documentation set.\n- [`doca-version`](..\u002F..\u002Fdoca-version\u002FSKILL.md) — canonical\n  DOCA version-handling rules. The\n  [`## Version compatibility`](CAPABILITIES.md#version-compatibility)\n  section in this skill is a thin overlay on top.\n- [`doca-setup`](..\u002F..\u002Fdoca-setup\u002FSKILL.md) — env preparation,\n  install verification, hugepages, NUMA awareness, and the\n  *I have no install yet* path with the public NGC DOCA\n  container.\n- [`doca-debug`](..\u002F..\u002Fdoca-debug\u002FSKILL.md) — the cross-cutting\n  debug ladder. DPA-perf surfaces *its own* error taxonomy;\n  when the cause turns out to be below DOCA, the taxonomy\n  hands off to `doca-debug`.\n- [`doca-hardware-safety`](..\u002F..\u002Fdoca-hardware-safety\u002FSKILL.md) —\n  the cross-cutting hardware-safety meta-policy this skill's\n  `## Safety policy` overlays.\n",{"data":35,"body":39},{"license":26,"name":4,"description":6,"metadata":36,"compatibility":38},{"kind":37},"tool","Requires DOCA SDK installed at \u002Fopt\u002Fmellanox\u002Fdoca on Linux (Ubuntu 22.04\u002F24.04 or RHEL\u002FSLES) with a DPA-capable device attached — ConnectX-7 as the minimum supported ConnectX generation, ConnectX-8 recommended, or BlueField-3 (BlueField-2 and earlier ConnectX are unsupported). VNF Flow mode required; PF or VF only (SFs are not supported on the DPA path). Reads `pkg-config doca-flow` and the shipped `doca_flow_dpa_perf` binary plus its README on the user's install.\n",{"type":40,"children":41},"root",[42,60,198,205,217,443,449,461,506,586,611,617,675,681,693,726,776,782,794,929,934,940,952,1047,1053,1159,1165],{"type":43,"tag":44,"props":45,"children":47},"element","h1",{"id":46},"doca-flow-dpa-perf-doca_flow_dpa_perf",[48,51,58],{"type":49,"value":50},"text","DOCA Flow DPA Perf (",{"type":43,"tag":52,"props":53,"children":55},"code",{"className":54},[],[56],{"type":49,"value":57},"doca_flow_dpa_perf",{"type":49,"value":59},")",{"type":43,"tag":61,"props":62,"children":63},"p",{},[64,70,72,77,79,89,91,101,103,113,115,125,127,136,138,151,153,158,160,165,167,172,174,184,186,196],{"type":43,"tag":65,"props":66,"children":67},"strong",{},[68],{"type":49,"value":69},"Where to start:",{"type":49,"value":71}," This is a tool skill for invoking\n",{"type":43,"tag":52,"props":73,"children":75},{"className":74},[],[76],{"type":49,"value":57},{"type":49,"value":78},", the DPA-accelerated Flow performance tool.\nOpen ",{"type":43,"tag":80,"props":81,"children":83},"a",{"href":82},"TASKS.md",[84],{"type":43,"tag":52,"props":85,"children":87},{"className":86},[],[88],{"type":49,"value":82},{"type":49,"value":90}," and start at\n",{"type":43,"tag":80,"props":92,"children":94},{"href":93},"TASKS.md#configure",[95],{"type":43,"tag":52,"props":96,"children":98},{"className":97},[],[99],{"type":49,"value":100},"## configure",{"type":49,"value":102}," to confirm DPA-capable\nhardware + VNF Flow mode + the active \u002F passive device split, then\n",{"type":43,"tag":80,"props":104,"children":106},{"href":105},"TASKS.md#run",[107],{"type":43,"tag":52,"props":108,"children":110},{"className":109},[],[111],{"type":49,"value":112},"## run",{"type":49,"value":114}," for the smoke-before-bulk flow with a\nsmall operation count before any sweep, then\n",{"type":43,"tag":80,"props":116,"children":118},{"href":117},"TASKS.md#test",[119],{"type":43,"tag":52,"props":120,"children":122},{"className":121},[],[123],{"type":49,"value":124},"## test",{"type":49,"value":126}," for the eval-loop overlay that gates\ndefensible Kops\u002Fsec numbers. Open ",{"type":43,"tag":80,"props":128,"children":130},{"href":129},"CAPABILITIES.md",[131],{"type":43,"tag":52,"props":132,"children":134},{"className":133},[],[135],{"type":49,"value":129},{"type":49,"value":137},"\nwhen the question is ",{"type":43,"tag":139,"props":140,"children":141},"em",{},[142,144,149],{"type":49,"value":143},"what ",{"type":43,"tag":52,"props":145,"children":147},{"className":146},[],[148],{"type":49,"value":57},{"type":49,"value":150}," can measure",{"type":49,"value":152},",\n",{"type":43,"tag":139,"props":154,"children":155},{},[156],{"type":49,"value":157},"what the DPA preconditions are",{"type":49,"value":159},", ",{"type":43,"tag":139,"props":161,"children":162},{},[163],{"type":49,"value":164},"which devices it runs on",{"type":49,"value":166},",\nor ",{"type":43,"tag":139,"props":168,"children":169},{},[170],{"type":49,"value":171},"how to interpret update \u002F disable \u002F self-test output without\nfooling yourself",{"type":49,"value":173},". If DOCA is not installed yet, route to\n",{"type":43,"tag":80,"props":175,"children":177},{"href":176},"..\u002F..\u002Fdoca-setup\u002FSKILL.md",[178],{"type":43,"tag":52,"props":179,"children":181},{"className":180},[],[182],{"type":49,"value":183},"doca-setup",{"type":49,"value":185}," first; if the device is\nnot DPA-capable (no ConnectX-7+ or BlueField-3+) then this tool is\nthe wrong surface and the right answer is\n",{"type":43,"tag":80,"props":187,"children":189},{"href":188},"..\u002Fdoca-flow-perf\u002FSKILL.md",[190],{"type":43,"tag":52,"props":191,"children":193},{"className":192},[],[194],{"type":49,"value":195},"doca-flow-perf",{"type":49,"value":197},".",{"type":43,"tag":199,"props":200,"children":202},"h2",{"id":201},"example-questions-this-skill-answers-well",[203],{"type":49,"value":204},"Example questions this skill answers well",{"type":43,"tag":61,"props":206,"children":207},{},[208,210,215],{"type":49,"value":209},"The CLASSES of ",{"type":43,"tag":52,"props":211,"children":213},{"className":212},[],[214],{"type":49,"value":57},{"type":49,"value":216}," questions this skill is built\nto answer, each with one worked example. The class is the\nload-bearing piece; the worked example is one instance.",{"type":43,"tag":218,"props":219,"children":220},"ul",{},[221,273,299,326,364,409],{"type":43,"tag":222,"props":223,"children":224},"li",{},[225,230,232,252,254,259,261,271],{"type":43,"tag":65,"props":226,"children":227},{},[228],{"type":49,"value":229},"\"Should I measure the DPA-offloaded Flow path or the\nhost \u002F DPU-CPU Flow path for this question?\"",{"type":49,"value":231}," — worked\nexample: ",{"type":43,"tag":139,"props":233,"children":234},{},[235,237,242,244,250],{"type":49,"value":236},"\"my workload programs path-selector entries via\nDOCA Flow; do I baseline with ",{"type":43,"tag":52,"props":238,"children":240},{"className":239},[],[241],{"type":49,"value":57},{"type":49,"value":243}," or with\n",{"type":43,"tag":52,"props":245,"children":247},{"className":246},[],[248],{"type":49,"value":249},"doca_flow_perf",{"type":49,"value":251},"?\"",{"type":49,"value":253},". Answered by the ",{"type":43,"tag":139,"props":255,"children":256},{},[257],{"type":49,"value":258},"DPA-vs-host",{"type":49,"value":260}," boundary\nin\n",{"type":43,"tag":80,"props":262,"children":264},{"href":263},"CAPABILITIES.md#capabilities-and-modes",[265],{"type":43,"tag":52,"props":266,"children":268},{"className":267},[],[269],{"type":49,"value":270},"CAPABILITIES.md ## Capabilities and modes",{"type":49,"value":272},"\nand the device-preconditions table.",{"type":43,"tag":222,"props":274,"children":275},{},[276,281,283,288,290,298],{"type":43,"tag":65,"props":277,"children":278},{},[279],{"type":49,"value":280},"\"What does the DPA-offload actually accelerate, and what\ndoesn't it change?\"",{"type":49,"value":282}," — worked example: ",{"type":43,"tag":139,"props":284,"children":285},{},[286],{"type":49,"value":287},"\"if I move my Flow\nrule update path to the DPA, what changes in the data plane\nfor the packets themselves?\"",{"type":49,"value":289},". Answered by the DPA-Provider\nscope in\n",{"type":43,"tag":80,"props":291,"children":292},{"href":263},[293],{"type":43,"tag":52,"props":294,"children":296},{"className":295},[],[297],{"type":49,"value":270},{"type":49,"value":197},{"type":43,"tag":222,"props":300,"children":301},{},[302,307,309,314,316,324],{"type":43,"tag":65,"props":303,"children":304},{},[305],{"type":49,"value":306},"\"What hardware do I need to use this tool at all?\"",{"type":49,"value":308}," —\nworked example: ",{"type":43,"tag":139,"props":310,"children":311},{},[312],{"type":49,"value":313},"\"is my BlueField-2 DPA-capable?\"",{"type":49,"value":315},". Answered\nby the device-preconditions table in\n",{"type":43,"tag":80,"props":317,"children":318},{"href":263},[319],{"type":43,"tag":52,"props":320,"children":322},{"className":321},[],[323],{"type":49,"value":270},{"type":49,"value":325},"\n(BlueField-3 yes, BlueField-2 no; ConnectX-7 minimum\nsupported, ConnectX-8 recommended, and later generations\nsupported per the public guide and the\nshipped README on the user's install).",{"type":43,"tag":222,"props":327,"children":328},{},[329,334,335,340,342,351,353,363],{"type":43,"tag":65,"props":330,"children":331},{},[332],{"type":49,"value":333},"\"How do I size my run — burst, queue, completion threshold,\nnumber of operations, iterations — to get a defensible\nKops\u002Fsec number?\"",{"type":49,"value":282},{"type":43,"tag":139,"props":336,"children":337},{},[338],{"type":49,"value":339},"\"I want the median\niteration time and standard deviation, not a single noisy\nfirst-iteration spike\"",{"type":49,"value":341},". Answered by the eval-loop overlay\nin\n",{"type":43,"tag":80,"props":343,"children":344},{"href":117},[345],{"type":43,"tag":52,"props":346,"children":348},{"className":347},[],[349],{"type":49,"value":350},"TASKS.md ## test",{"type":49,"value":352}," and the iteration-stats\nrule in\n",{"type":43,"tag":80,"props":354,"children":356},{"href":355},"CAPABILITIES.md#observability",[357],{"type":43,"tag":52,"props":358,"children":360},{"className":359},[],[361],{"type":49,"value":362},"CAPABILITIES.md ## Observability",{"type":49,"value":197},{"type":43,"tag":222,"props":365,"children":366},{},[367,372,373,378,380,390],{"type":43,"tag":65,"props":368,"children":369},{},[370],{"type":49,"value":371},"\"My tool reports zero ops \u002F hangs \u002F fails the self-test —\nwhat does that mean?\"",{"type":49,"value":282},{"type":43,"tag":139,"props":374,"children":375},{},[376],{"type":49,"value":377},"\"the tool runs but\nthe self-test step fails\"",{"type":49,"value":379},". Answered by the layered error\ntaxonomy in\n",{"type":43,"tag":80,"props":381,"children":383},{"href":382},"CAPABILITIES.md#error-taxonomy",[384],{"type":43,"tag":52,"props":385,"children":387},{"className":386},[],[388],{"type":49,"value":389},"CAPABILITIES.md ## Error taxonomy",{"type":43,"tag":218,"props":391,"children":392},{},[393],{"type":43,"tag":222,"props":394,"children":395},{},[396,398,408],{"type":49,"value":397},"the debug ladder in\n",{"type":43,"tag":80,"props":399,"children":401},{"href":400},"TASKS.md#debug",[402],{"type":43,"tag":52,"props":403,"children":405},{"className":404},[],[406],{"type":49,"value":407},"TASKS.md ## debug",{"type":49,"value":197},{"type":43,"tag":222,"props":410,"children":411},{},[412,417,418,423,425,435],{"type":43,"tag":65,"props":413,"children":414},{},[415],{"type":49,"value":416},"\"How do I quote a DPA-perf number alongside a host-side\nFlow-perf number for the same workload, in a way the next\nengineer can actually compare?\"",{"type":49,"value":282},{"type":43,"tag":139,"props":419,"children":420},{},[421],{"type":49,"value":422},"\"two\nKops\u002Fsec numbers for what is supposedly the same workload\"",{"type":49,"value":424},".\nAnswered by the four-tuple capture rule in\n",{"type":43,"tag":80,"props":426,"children":428},{"href":427},"CAPABILITIES.md#safety-policy",[429],{"type":43,"tag":52,"props":430,"children":432},{"className":431},[],[433],{"type":49,"value":434},"CAPABILITIES.md ## Safety policy",{"type":43,"tag":218,"props":436,"children":437},{},[438],{"type":43,"tag":222,"props":439,"children":440},{},[441],{"type":49,"value":442},"the per-tool-name rule (the host tool and the DPA tool are\ndifferent surfaces; their numbers are not interchangeable\nwithout naming which tool produced which).",{"type":43,"tag":199,"props":444,"children":446},{"id":445},"audience",[447],{"type":49,"value":448},"Audience",{"type":43,"tag":61,"props":450,"children":451},{},[452,454,459],{"type":49,"value":453},"This skill serves ",{"type":43,"tag":65,"props":455,"children":456},{},[457],{"type":49,"value":458},"external operators, performance engineers,\nDOCA Flow application developers, and AI agents who need a\ndefensible measurement of the DPA-offloaded Flow update path",{"type":49,"value":460}," on\nDPA-capable hardware. Concretely:",{"type":43,"tag":218,"props":462,"children":463},{},[464,469,481,494],{"type":43,"tag":222,"props":465,"children":466},{},[467],{"type":49,"value":468},"A platform operator deciding whether to move a path-selector\nworkload onto the DPA versus keeping it on the host \u002F DPU-CPU\npath, and wanting a number to compare.",{"type":43,"tag":222,"props":470,"children":471},{},[472,474,479],{"type":49,"value":473},"A performance engineer producing a ",{"type":43,"tag":139,"props":475,"children":476},{},[477],{"type":49,"value":478},"\"DPA Kops\u002Fsec for update\noperation, queue-size X, burst-size Y, N workers\"",{"type":49,"value":480}," baseline\non a specific device + DOCA version so a downstream\ncomparison is meaningful.",{"type":43,"tag":222,"props":482,"children":483},{},[484,486,492],{"type":49,"value":485},"A DOCA Flow application developer who has already used\n",{"type":43,"tag":52,"props":487,"children":489},{"className":488},[],[490],{"type":49,"value":491},"doca-dpa",{"type":49,"value":493}," to land a DPA-offload of their Flow rule update\npath and wants to characterize what the device delivers.",{"type":43,"tag":222,"props":495,"children":496},{},[497,499,504],{"type":49,"value":498},"An AI agent answering ",{"type":43,"tag":139,"props":500,"children":501},{},[502],{"type":49,"value":503},"\"what update rate should I expect from\nthe DPA-offloaded Flow path on device Y?\"",{"type":49,"value":505}," honestly — with a\nmeasured number, the command line that produced it, and the\ndevice + DOCA version + as-deployed environment that scopes\nit — instead of guessing from datasheet headlines.",{"type":43,"tag":61,"props":507,"children":508},{},[509,511,516,518,522,524,530,531,535,537,543,545,550,552,561,563,568,570,574,576,584],{"type":49,"value":510},"It is ",{"type":43,"tag":65,"props":512,"children":513},{},[514],{"type":49,"value":515},"not",{"type":49,"value":517}," for users debugging the tool's source code,\n",{"type":43,"tag":65,"props":519,"children":520},{},[521],{"type":49,"value":515},{"type":49,"value":523}," a substitute for the live public DOCA Flow DPA Perf guide\non ",{"type":43,"tag":52,"props":525,"children":527},{"className":526},[],[528],{"type":49,"value":529},"docs.nvidia.com",{"type":49,"value":159},{"type":43,"tag":65,"props":532,"children":533},{},[534],{"type":49,"value":515},{"type":49,"value":536}," the place to learn the ",{"type":43,"tag":52,"props":538,"children":540},{"className":539},[],[541],{"type":49,"value":542},"doca-flow",{"type":49,"value":544},"\nor ",{"type":43,"tag":52,"props":546,"children":548},{"className":547},[],[549],{"type":49,"value":491},{"type":49,"value":551}," APIs (that audience belongs in\n",{"type":43,"tag":80,"props":553,"children":555},{"href":554},"..\u002F..\u002Flibs\u002Fdoca-flow\u002FSKILL.md",[556],{"type":43,"tag":52,"props":557,"children":559},{"className":558},[],[560],{"type":49,"value":542},{"type":49,"value":562}," and\n",{"type":43,"tag":52,"props":564,"children":566},{"className":565},[],[567],{"type":49,"value":491},{"type":49,"value":569},"), and ",{"type":43,"tag":65,"props":571,"children":572},{},[573],{"type":49,"value":515},{"type":49,"value":575}," the right\ntool for the host \u002F DPU-CPU Flow path (route to\n",{"type":43,"tag":80,"props":577,"children":578},{"href":188},[579],{"type":43,"tag":52,"props":580,"children":582},{"className":581},[],[583],{"type":49,"value":195},{"type":49,"value":585},").",{"type":43,"tag":61,"props":587,"children":588},{},[589,594,596,601,603,609],{"type":43,"tag":52,"props":590,"children":592},{"className":591},[],[593],{"type":49,"value":57},{"type":49,"value":595}," is shipped as a ",{"type":43,"tag":65,"props":597,"children":598},{},[599],{"type":49,"value":600},"single CLI binary",{"type":49,"value":602}," with\nDPA-side device code linked in. The skill uses the same\n",{"type":43,"tag":52,"props":604,"children":606},{"className":605},[],[607],{"type":49,"value":608},"kind: tool",{"type":49,"value":610}," three-file shape as the rest of the bundle so\nthe agent's task-verb contract is uniform across the bundle.",{"type":43,"tag":199,"props":612,"children":614},{"id":613},"language-scope",[615],{"type":49,"value":616},"Language scope",{"type":43,"tag":61,"props":618,"children":619},{},[620,622,627,629,635,637,643,645,650,652,657,659,664,666,674],{"type":49,"value":621},"This skill governs invocation, output interpretation, and\nrecommendation-of-routing for the ",{"type":43,"tag":52,"props":623,"children":625},{"className":624},[],[626],{"type":49,"value":57},{"type":49,"value":628}," CLI on\nDPA-capable hardware. The tool itself has both a host-side\ncontrol (C-language ARGP + DOCA + DPDK code per the shipped\n",{"type":43,"tag":52,"props":630,"children":632},{"className":631},[],[633],{"type":49,"value":634},"flow_dpa_perf.c",{"type":49,"value":636}," \u002F ",{"type":43,"tag":52,"props":638,"children":640},{"className":639},[],[641],{"type":49,"value":642},"flow_dpa_perf_core.c",{"type":49,"value":644},") and a DPA-side device\ncomponent (DPA-side code on the shipped DPA device runtime).\nExternal users do not link any of this; what they configure is\nthe JSON-config-or-CLI invocation surface. For the\n",{"type":43,"tag":52,"props":646,"children":648},{"className":647},[],[649],{"type":49,"value":491},{"type":49,"value":651}," programming model behind the DPA-side execution\nengine, see\n",{"type":43,"tag":52,"props":653,"children":655},{"className":654},[],[656],{"type":49,"value":491},{"type":49,"value":658},"; for the ",{"type":43,"tag":52,"props":660,"children":662},{"className":661},[],[663],{"type":49,"value":542},{"type":49,"value":665},"\nAPI behind the pipeline the DPA path executes, see\n",{"type":43,"tag":80,"props":667,"children":668},{"href":554},[669],{"type":43,"tag":52,"props":670,"children":672},{"className":671},[],[673],{"type":49,"value":542},{"type":49,"value":197},{"type":43,"tag":199,"props":676,"children":678},{"id":677},"when-to-load-this-skill",[679],{"type":49,"value":680},"When to load this skill",{"type":43,"tag":61,"props":682,"children":683},{},[684,686,691],{"type":49,"value":685},"Load this skill when the user is — or the agent needs to —\ninvoke ",{"type":43,"tag":52,"props":687,"children":689},{"className":688},[],[690],{"type":49,"value":57},{"type":49,"value":692}," on a real host with DOCA installed\nand a DPA-capable device attached (or the public NGC DOCA\ncontainer with the equivalent device passthrough) to measure\nupdate \u002F disable rates on the DPA-offloaded Flow path.\nConcretely:",{"type":43,"tag":218,"props":694,"children":695},{},[696,701,706,711,716,721],{"type":43,"tag":222,"props":697,"children":698},{},[699],{"type":49,"value":700},"Confirming DPA preconditions (DPA-capable device class,\nVNF Flow mode, recommended PF use, no SFs) before invoking\nthe tool.",{"type":43,"tag":222,"props":702,"children":703},{},[704],{"type":49,"value":705},"Picking the active \u002F passive device split appropriate to the\nuser's hardware (two-port BlueField-3 active + passive; one-\nport ConnectX-9 active only).",{"type":43,"tag":222,"props":707,"children":708},{},[709],{"type":49,"value":710},"Picking the workload-shape axes (burst size, queue size,\ncompletion threshold, hash pipe algorithm, work policy,\nnumber of PSL tables, table size, number of workers).",{"type":43,"tag":222,"props":712,"children":713},{},[714],{"type":49,"value":715},"Picking the operation axis (update or disable-enable) per the\nshipped README's documented operations.",{"type":43,"tag":222,"props":717,"children":718},{},[719],{"type":49,"value":720},"Producing a defensible Kops\u002Fsec number with iteration stats\n(median, max, standard deviation) captured.",{"type":43,"tag":222,"props":722,"children":723},{},[724],{"type":49,"value":725},"Diagnosing zero-ops \u002F hung \u002F failed-self-test runs through\nthe layered error taxonomy.",{"type":43,"tag":61,"props":727,"children":728},{},[729,731,735,737,747,749,755,757,765,767,775],{"type":49,"value":730},"Do ",{"type":43,"tag":65,"props":732,"children":733},{},[734],{"type":49,"value":515},{"type":49,"value":736}," load this skill for general DOCA orientation, Flow\nprogram API work, or installation. For those, use\n",{"type":43,"tag":80,"props":738,"children":740},{"href":739},"..\u002F..\u002Fdoca-public-knowledge-map\u002FSKILL.md",[741],{"type":43,"tag":52,"props":742,"children":744},{"className":743},[],[745],{"type":49,"value":746},"doca-public-knowledge-map",{"type":49,"value":748},",\nthe matching ",{"type":43,"tag":52,"props":750,"children":752},{"className":751},[],[753],{"type":49,"value":754},"libs\u002F\u003Clibrary>",{"type":49,"value":756}," skill, or\n",{"type":43,"tag":80,"props":758,"children":759},{"href":176},[760],{"type":43,"tag":52,"props":761,"children":763},{"className":762},[],[764],{"type":49,"value":183},{"type":49,"value":766},". Do not load it for\nthe host \u002F DPU-CPU Flow path — that audience belongs in\n",{"type":43,"tag":80,"props":768,"children":769},{"href":188},[770],{"type":43,"tag":52,"props":771,"children":773},{"className":772},[],[774],{"type":49,"value":195},{"type":49,"value":197},{"type":43,"tag":199,"props":777,"children":779},{"id":778},"what-this-skill-provides",[780],{"type":49,"value":781},"What this skill provides",{"type":43,"tag":61,"props":783,"children":784},{},[785,787,792],{"type":49,"value":786},"This is a ",{"type":43,"tag":65,"props":788,"children":789},{},[790],{"type":49,"value":791},"thin loader",{"type":49,"value":793},". Substantive material lives in two\ncompanion files:",{"type":43,"tag":218,"props":795,"children":796},{},[797,840],{"type":43,"tag":222,"props":798,"children":799},{},[800,805,807,812,814,819,821,826,828,838],{"type":43,"tag":52,"props":801,"children":803},{"className":802},[],[804],{"type":49,"value":129},{"type":49,"value":806}," — what ",{"type":43,"tag":52,"props":808,"children":810},{"className":809},[],[811],{"type":49,"value":57},{"type":49,"value":813}," measures\n(the DPA-Provider-on-DPA-device update \u002F disable path\nspecifically), the DPA-vs-host-path boundary, the\ndevice-preconditions table (ConnectX-7+ \u002F BlueField-3+),\nthe documented VNF-only Flow-mode rule, the PF-vs-VF-vs-SF\nrule (SFs not supported on DPA), the workload-shape axes\n(burst, queue, completion threshold, hash pipe algorithm,\nwork policy, PSL tables, table size, workers), the\noperation axis (update vs disable-enable), the version\noverlay (this tool rides the ",{"type":43,"tag":52,"props":815,"children":817},{"className":816},[],[818],{"type":49,"value":542},{"type":49,"value":820}," and ",{"type":43,"tag":52,"props":822,"children":824},{"className":823},[],[825],{"type":49,"value":491},{"type":49,"value":827},"\nversions it links against; the canonical rules live in\n",{"type":43,"tag":80,"props":829,"children":831},{"href":830},"..\u002F..\u002Fdoca-version\u002FSKILL.md",[832],{"type":43,"tag":52,"props":833,"children":835},{"className":834},[],[836],{"type":49,"value":837},"doca-version",{"type":49,"value":839},"), the layered\nerror taxonomy\n(config-syntax \u002F device-binding \u002F dpa-precondition \u002F\nworkload-precondition \u002F measurement-soundness \u002F self-test \u002F\nversion \u002F cross-cutting), the observability surface\n(iteration statistics, self-test path-selector verification,\ntcpdump-side traffic verification), and the safety posture\n(smoke-before-bulk, four-tuple capture, name the tool that\nproduced the number).",{"type":43,"tag":222,"props":841,"children":842},{},[843,848,850,856,858,864,866,872,874,880,882,888,890,896,898,904,906,912,914,920,922,928],{"type":43,"tag":52,"props":844,"children":846},{"className":845},[],[847],{"type":49,"value":82},{"type":49,"value":849}," — step-by-step workflows for the in-scope task\nverbs: ",{"type":43,"tag":52,"props":851,"children":853},{"className":852},[],[854],{"type":49,"value":855},"install",{"type":49,"value":857}," (route to setup; the binary is shipped),\n",{"type":43,"tag":52,"props":859,"children":861},{"className":860},[],[862],{"type":49,"value":863},"configure",{"type":49,"value":865}," (DPA-preconditions + active \u002F passive device +\nworkload-shape decision), ",{"type":43,"tag":52,"props":867,"children":869},{"className":868},[],[870],{"type":49,"value":871},"build",{"type":49,"value":873}," (route to install — the\nbinary is shipped), ",{"type":43,"tag":52,"props":875,"children":877},{"className":876},[],[878],{"type":49,"value":879},"modify",{"type":49,"value":881}," (refuse — modify the invocation,\nnot the binary), ",{"type":43,"tag":52,"props":883,"children":885},{"className":884},[],[886],{"type":49,"value":887},"run",{"type":49,"value":889}," (smoke before bulk), ",{"type":43,"tag":52,"props":891,"children":893},{"className":892},[],[894],{"type":49,"value":895},"test",{"type":49,"value":897}," (eval\nloop), ",{"type":43,"tag":52,"props":899,"children":901},{"className":900},[],[902],{"type":49,"value":903},"debug",{"type":49,"value":905}," (layered diagnosis), ",{"type":43,"tag":52,"props":907,"children":909},{"className":908},[],[910],{"type":49,"value":911},"use",{"type":49,"value":913}," (consume the\ncaptured number), plus a ",{"type":43,"tag":52,"props":915,"children":917},{"className":916},[],[918],{"type":49,"value":919},"Deferred task verbs",{"type":49,"value":921}," block routing\nout-of-scope questions and a ",{"type":43,"tag":52,"props":923,"children":925},{"className":924},[],[926],{"type":49,"value":927},"Command appendix",{"type":49,"value":197},{"type":43,"tag":61,"props":930,"children":931},{},[932],{"type":49,"value":933},"The skill assumes a host where DOCA is already installed (or\nthe NGC DOCA container is running) on a DPA-capable device and\nthe operator has the permissions to bind the device and allocate\nthe DPA execution resources the tool needs.",{"type":43,"tag":199,"props":935,"children":937},{"id":936},"what-this-skill-deliberately-does-not-ship",[938],{"type":49,"value":939},"What this skill deliberately does not ship",{"type":43,"tag":61,"props":941,"children":942},{},[943,945,950],{"type":49,"value":944},"This skill is ",{"type":43,"tag":65,"props":946,"children":947},{},[948],{"type":49,"value":949},"agent guidance",{"type":49,"value":951},", not a samples or scripts\nbundle. To keep the boundary clean, it deliberately does not\ncontain — and pull requests should not add:",{"type":43,"tag":218,"props":953,"children":954},{},[955,987,1004,1014],{"type":43,"tag":222,"props":956,"children":957},{},[958,971,973,978,980,985],{"type":43,"tag":65,"props":959,"children":960},{},[961,963,969],{"type":49,"value":962},"Verbatim default values for flag inventories beyond what\nthe shipped README or installed ",{"type":43,"tag":52,"props":964,"children":966},{"className":965},[],[967],{"type":49,"value":968},"--help",{"type":49,"value":970}," documents.",{"type":49,"value":972}," Read\ndefaults from the README first, then fall back to the\ninstalled binary's ",{"type":43,"tag":52,"props":974,"children":976},{"className":975},[],[977],{"type":49,"value":968},{"type":49,"value":979},". If neither defines a needed\ndefault, stop and request the operator's explicit value\ninstead of guessing. The\nflag surface is install-specific within the documented\nsurface; the documented invocations + ",{"type":43,"tag":52,"props":981,"children":983},{"className":982},[],[984],{"type":49,"value":968},{"type":49,"value":986}," on the\ninstalled version are the authoritative answer. Inventing\na flag is the most common hallucination failure.",{"type":43,"tag":222,"props":988,"children":989},{},[990,995,997,1002],{"type":43,"tag":65,"props":991,"children":992},{},[993],{"type":49,"value":994},"Pre-baked example Kops\u002Fsec numbers or expected throughput\nnumbers.",{"type":49,"value":996}," Output is device-, firmware-, DOCA-version-,\nworkload-, and platform-specific; a pinned number for one\nplatform misleads operators on a different platform \u002F\nversion. The shipped README's example numbers are\n",{"type":43,"tag":139,"props":998,"children":999},{},[1000],{"type":49,"value":1001},"illustrative",{"type":49,"value":1003},", not a baseline the agent should quote as\nground truth.",{"type":43,"tag":222,"props":1005,"children":1006},{},[1007,1012],{"type":43,"tag":65,"props":1008,"children":1009},{},[1010],{"type":49,"value":1011},"Wrappers, parsers, or scripts",{"type":49,"value":1013}," in any language that\nconsume the tool's stdout \u002F CSV. The output format is\ndocumented; if a user wants to script against it, the\nright answer is \"read the live guide, write the parser\nagainst your installed version\".",{"type":43,"tag":222,"props":1015,"children":1016},{},[1017,1038,1040,1045],{"type":43,"tag":65,"props":1018,"children":1019},{},[1020,1022,1028,1030,1036],{"type":49,"value":1021},"A ",{"type":43,"tag":52,"props":1023,"children":1025},{"className":1024},[],[1026],{"type":49,"value":1027},"samples\u002F",{"type":49,"value":1029}," or ",{"type":43,"tag":52,"props":1031,"children":1033},{"className":1032},[],[1034],{"type":49,"value":1035},"reference\u002F",{"type":49,"value":1037}," subtree.",{"type":49,"value":1039}," This is a thin\nloader for a documented CLI; substantive material lives on\nthe public page, in ",{"type":43,"tag":52,"props":1041,"children":1043},{"className":1042},[],[1044],{"type":49,"value":968},{"type":49,"value":1046},", and in the shipped README on\nthe user's install.",{"type":43,"tag":199,"props":1048,"children":1050},{"id":1049},"loading-order",[1051],{"type":49,"value":1052},"Loading order",{"type":43,"tag":1054,"props":1055,"children":1056},"ol",{},[1057,1077,1097],{"type":43,"tag":222,"props":1058,"children":1059},{},[1060,1062,1068,1070,1075],{"type":49,"value":1061},"Read this ",{"type":43,"tag":52,"props":1063,"children":1065},{"className":1064},[],[1066],{"type":49,"value":1067},"SKILL.md",{"type":49,"value":1069}," first to confirm the user's question\nis in scope (the user actually wants to invoke\n",{"type":43,"tag":52,"props":1071,"children":1073},{"className":1072},[],[1074],{"type":49,"value":57},{"type":49,"value":1076}," on DPA-capable hardware, not measure\nthe host \u002F DPU-CPU Flow path).",{"type":43,"tag":222,"props":1078,"children":1079},{},[1080],{"type":43,"tag":65,"props":1081,"children":1082},{},[1083,1085,1090,1092,1096],{"type":49,"value":1084},"For what ",{"type":43,"tag":52,"props":1086,"children":1088},{"className":1087},[],[1089],{"type":49,"value":57},{"type":49,"value":1091}," measures, the DPA-vs-host\nboundary, the device-preconditions table, the workload-\nshape axes, the version overlay, the error taxonomy, the\nobservability surface, and the safety posture, see\n",{"type":43,"tag":80,"props":1093,"children":1094},{"href":129},[1095],{"type":49,"value":129},{"type":49,"value":197},{"type":43,"tag":222,"props":1098,"children":1099},{},[1100],{"type":43,"tag":65,"props":1101,"children":1102},{},[1103,1105,1110,1111,1116,1117,1122,1123,1128,1129,1134,1135,1140,1141,1146,1147,1152,1154,1158],{"type":49,"value":1104},"For the documented invocations and the smoke-before-bulk\nworkflow — ",{"type":43,"tag":52,"props":1106,"children":1108},{"className":1107},[],[1109],{"type":49,"value":855},{"type":49,"value":159},{"type":43,"tag":52,"props":1112,"children":1114},{"className":1113},[],[1115],{"type":49,"value":863},{"type":49,"value":159},{"type":43,"tag":52,"props":1118,"children":1120},{"className":1119},[],[1121],{"type":49,"value":871},{"type":49,"value":159},{"type":43,"tag":52,"props":1124,"children":1126},{"className":1125},[],[1127],{"type":49,"value":879},{"type":49,"value":152},{"type":43,"tag":52,"props":1130,"children":1132},{"className":1131},[],[1133],{"type":49,"value":887},{"type":49,"value":159},{"type":43,"tag":52,"props":1136,"children":1138},{"className":1137},[],[1139],{"type":49,"value":895},{"type":49,"value":159},{"type":43,"tag":52,"props":1142,"children":1144},{"className":1143},[],[1145],{"type":49,"value":903},{"type":49,"value":159},{"type":43,"tag":52,"props":1148,"children":1150},{"className":1149},[],[1151],{"type":49,"value":911},{"type":49,"value":1153}," — see ",{"type":43,"tag":80,"props":1155,"children":1156},{"href":82},[1157],{"type":49,"value":82},{"type":49,"value":197},{"type":43,"tag":199,"props":1160,"children":1162},{"id":1161},"related-skills",[1163],{"type":49,"value":1164},"Related skills",{"type":43,"tag":218,"props":1166,"children":1167},{},[1168,1195,1219,1242,1264,1284,1309,1329,1357],{"type":43,"tag":222,"props":1169,"children":1170},{},[1171,1179,1181,1186,1188,1193],{"type":43,"tag":80,"props":1172,"children":1173},{"href":554},[1174],{"type":43,"tag":52,"props":1175,"children":1177},{"className":1176},[],[1178],{"type":49,"value":542},{"type":49,"value":1180}," — the ",{"type":43,"tag":65,"props":1182,"children":1183},{},[1184],{"type":49,"value":1185},"base\nlibrary",{"type":49,"value":1187}," whose pipeline this tool measures on the DPA\npath. The pipe \u002F entry \u002F rule surface this tool drives is\ncreated by ",{"type":43,"tag":52,"props":1189,"children":1191},{"className":1190},[],[1192],{"type":49,"value":542},{"type":49,"value":1194}," program code; the library's pipe\nattributes and capability surface are the upstream context.",{"type":43,"tag":222,"props":1196,"children":1197},{},[1198,1203,1205,1210,1212,1217],{"type":43,"tag":52,"props":1199,"children":1201},{"className":1200},[],[1202],{"type":49,"value":491},{"type":49,"value":1204}," — the\nprogramming model behind the DPA execution engine the tool\nruns on. When the user's question goes from ",{"type":43,"tag":139,"props":1206,"children":1207},{},[1208],{"type":49,"value":1209},"\"measure the\nDPA path\"",{"type":49,"value":1211}," to ",{"type":43,"tag":139,"props":1213,"children":1214},{},[1215],{"type":49,"value":1216},"\"why is the DPA path doing this\"",{"type":49,"value":1218},", that\nskill is the next stop.",{"type":43,"tag":222,"props":1220,"children":1221},{},[1222,1230,1232,1240],{"type":43,"tag":80,"props":1223,"children":1224},{"href":188},[1225],{"type":43,"tag":52,"props":1226,"children":1228},{"className":1227},[],[1229],{"type":49,"value":195},{"type":49,"value":1231}," — the\nhost \u002F DPU-CPU Flow performance tool. The cross-tool\ncomparison rule lives in\n",{"type":43,"tag":80,"props":1233,"children":1234},{"href":263},[1235],{"type":43,"tag":52,"props":1236,"children":1238},{"className":1237},[],[1239],{"type":49,"value":270},{"type":49,"value":1241},":\nname which tool produced which number.",{"type":43,"tag":222,"props":1243,"children":1244},{},[1245,1255,1257,1262],{"type":43,"tag":80,"props":1246,"children":1248},{"href":1247},"..\u002Fdoca-flow-tune\u002FSKILL.md",[1249],{"type":43,"tag":52,"props":1250,"children":1252},{"className":1251},[],[1253],{"type":49,"value":1254},"doca-flow-tune",{"type":49,"value":1256}," — the Flow\ntuning tool. A DPA-perf number is the kind of baseline\n",{"type":43,"tag":52,"props":1258,"children":1260},{"className":1259},[],[1261],{"type":49,"value":1254},{"type":49,"value":1263}," then optimizes on top of, via a Flow-program\nmodify-a-sample loop.",{"type":43,"tag":222,"props":1265,"children":1266},{},[1267,1275,1277,1282],{"type":43,"tag":80,"props":1268,"children":1269},{"href":739},[1270],{"type":43,"tag":52,"props":1271,"children":1273},{"className":1272},[],[1274],{"type":49,"value":746},{"type":49,"value":1276}," —\nrouting to the public DOCA Flow DPA Perf page on\n",{"type":43,"tag":52,"props":1278,"children":1280},{"className":1279},[],[1281],{"type":49,"value":529},{"type":49,"value":1283}," and the rest of the public DOCA\ndocumentation set.",{"type":43,"tag":222,"props":1285,"children":1286},{},[1287,1295,1297,1307],{"type":43,"tag":80,"props":1288,"children":1289},{"href":830},[1290],{"type":43,"tag":52,"props":1291,"children":1293},{"className":1292},[],[1294],{"type":49,"value":837},{"type":49,"value":1296}," — canonical\nDOCA version-handling rules. The\n",{"type":43,"tag":80,"props":1298,"children":1300},{"href":1299},"CAPABILITIES.md#version-compatibility",[1301],{"type":43,"tag":52,"props":1302,"children":1304},{"className":1303},[],[1305],{"type":49,"value":1306},"## Version compatibility",{"type":49,"value":1308},"\nsection in this skill is a thin overlay on top.",{"type":43,"tag":222,"props":1310,"children":1311},{},[1312,1320,1322,1327],{"type":43,"tag":80,"props":1313,"children":1314},{"href":176},[1315],{"type":43,"tag":52,"props":1316,"children":1318},{"className":1317},[],[1319],{"type":49,"value":183},{"type":49,"value":1321}," — env preparation,\ninstall verification, hugepages, NUMA awareness, and the\n",{"type":43,"tag":139,"props":1323,"children":1324},{},[1325],{"type":49,"value":1326},"I have no install yet",{"type":49,"value":1328}," path with the public NGC DOCA\ncontainer.",{"type":43,"tag":222,"props":1330,"children":1331},{},[1332,1342,1344,1349,1351,1356],{"type":43,"tag":80,"props":1333,"children":1335},{"href":1334},"..\u002F..\u002Fdoca-debug\u002FSKILL.md",[1336],{"type":43,"tag":52,"props":1337,"children":1339},{"className":1338},[],[1340],{"type":49,"value":1341},"doca-debug",{"type":49,"value":1343}," — the cross-cutting\ndebug ladder. DPA-perf surfaces ",{"type":43,"tag":139,"props":1345,"children":1346},{},[1347],{"type":49,"value":1348},"its own",{"type":49,"value":1350}," error taxonomy;\nwhen the cause turns out to be below DOCA, the taxonomy\nhands off to ",{"type":43,"tag":52,"props":1352,"children":1354},{"className":1353},[],[1355],{"type":49,"value":1341},{"type":49,"value":197},{"type":43,"tag":222,"props":1358,"children":1359},{},[1360,1370,1372,1378],{"type":43,"tag":80,"props":1361,"children":1363},{"href":1362},"..\u002F..\u002Fdoca-hardware-safety\u002FSKILL.md",[1364],{"type":43,"tag":52,"props":1365,"children":1367},{"className":1366},[],[1368],{"type":49,"value":1369},"doca-hardware-safety",{"type":49,"value":1371}," —\nthe cross-cutting hardware-safety meta-policy this skill's\n",{"type":43,"tag":52,"props":1373,"children":1375},{"className":1374},[],[1376],{"type":49,"value":1377},"## Safety policy",{"type":49,"value":1379}," overlays.",{"items":1381,"total":1540},[1382,1400,1418,1429,1441,1455,1468,1482,1495,1506,1520,1529],{"slug":1383,"name":1383,"fn":1384,"description":1385,"org":1386,"tags":1387,"stars":1397,"repoUrl":1398,"updatedAt":1399},"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},[1388,1391,1394],{"name":1389,"slug":1390,"type":15},"Documentation","documentation",{"name":1392,"slug":1393,"type":15},"MCP","mcp",{"name":1395,"slug":1396,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":1401,"name":1401,"fn":1402,"description":1403,"org":1404,"tags":1405,"stars":1415,"repoUrl":1416,"updatedAt":1417},"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},[1406,1409,1412],{"name":1407,"slug":1408,"type":15},"Containers","containers",{"name":1410,"slug":1411,"type":15},"Deployment","deployment",{"name":1413,"slug":1414,"type":15},"Python","python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":1419,"name":1419,"fn":1420,"description":1421,"org":1422,"tags":1423,"stars":1415,"repoUrl":1416,"updatedAt":1428},"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},[1424,1427],{"name":1425,"slug":1426,"type":15},"CI\u002FCD","ci-cd",{"name":1410,"slug":1411,"type":15},"2026-07-14T05:25:59.97109",{"slug":1430,"name":1430,"fn":1431,"description":1432,"org":1433,"tags":1434,"stars":1415,"repoUrl":1416,"updatedAt":1440},"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},[1435,1436,1437],{"name":1425,"slug":1426,"type":15},{"name":1410,"slug":1411,"type":15},{"name":1438,"slug":1439,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":1442,"name":1442,"fn":1443,"description":1444,"org":1445,"tags":1446,"stars":1415,"repoUrl":1416,"updatedAt":1454},"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},[1447,1450,1451],{"name":1448,"slug":1449,"type":15},"Debugging","debugging",{"name":1438,"slug":1439,"type":15},{"name":1452,"slug":1453,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":1456,"name":1456,"fn":1457,"description":1458,"org":1459,"tags":1460,"stars":1415,"repoUrl":1416,"updatedAt":1467},"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},[1461,1464],{"name":1462,"slug":1463,"type":15},"Best Practices","best-practices",{"name":1465,"slug":1466,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":1469,"name":1469,"fn":1470,"description":1471,"org":1472,"tags":1473,"stars":1415,"repoUrl":1416,"updatedAt":1481},"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},[1474,1477,1480],{"name":1475,"slug":1476,"type":15},"Machine Learning","machine-learning",{"name":1478,"slug":1479,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":1483,"name":1483,"fn":1484,"description":1485,"org":1486,"tags":1487,"stars":1415,"repoUrl":1416,"updatedAt":1494},"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},[1488,1491],{"name":1489,"slug":1490,"type":15},"QA","qa",{"name":1492,"slug":1493,"type":15},"Testing","testing","2026-07-14T05:25:53.673039",{"slug":1496,"name":1496,"fn":1497,"description":1498,"org":1499,"tags":1500,"stars":1415,"repoUrl":1416,"updatedAt":1505},"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},[1501,1502],{"name":1410,"slug":1411,"type":15},{"name":1503,"slug":1504,"type":15},"Infrastructure","infrastructure","2026-07-14T05:25:49.362534",{"slug":1507,"name":1507,"fn":1508,"description":1509,"org":1510,"tags":1511,"stars":1415,"repoUrl":1416,"updatedAt":1519},"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},[1512,1515,1516],{"name":1513,"slug":1514,"type":15},"Code Review","code-review",{"name":1438,"slug":1439,"type":15},{"name":1517,"slug":1518,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":1521,"name":1521,"fn":1522,"description":1523,"org":1524,"tags":1525,"stars":1415,"repoUrl":1416,"updatedAt":1528},"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},[1526,1527],{"name":1489,"slug":1490,"type":15},{"name":1492,"slug":1493,"type":15},"2026-07-14T05:25:54.928983",{"slug":1530,"name":1530,"fn":1531,"description":1532,"org":1533,"tags":1534,"stars":1415,"repoUrl":1416,"updatedAt":1539},"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},[1535,1538],{"name":1536,"slug":1537,"type":15},"Automation","automation",{"name":1425,"slug":1426,"type":15},"2026-07-30T05:29:03.275638",496,{"items":1542,"total":1636},[1543,1558,1568,1582,1592,1607,1622],{"slug":1544,"name":1544,"fn":1545,"description":1546,"org":1547,"tags":1548,"stars":23,"repoUrl":24,"updatedAt":1557},"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},[1549,1552,1555,1556],{"name":1550,"slug":1551,"type":15},"Data Analysis","data-analysis",{"name":1553,"slug":1554,"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":1559,"name":1559,"fn":1560,"description":1561,"org":1562,"tags":1563,"stars":23,"repoUrl":24,"updatedAt":1567},"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},[1564,1565,1566],{"name":1410,"slug":1411,"type":15},{"name":1503,"slug":1504,"type":15},{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":1569,"name":1569,"fn":1570,"description":1571,"org":1572,"tags":1573,"stars":23,"repoUrl":24,"updatedAt":1581},"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},[1574,1577,1578],{"name":1575,"slug":1576,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":1579,"slug":1580,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":1583,"name":1583,"fn":1584,"description":1585,"org":1586,"tags":1587,"stars":23,"repoUrl":24,"updatedAt":1591},"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},[1588,1589,1590],{"name":1550,"slug":1551,"type":15},{"name":9,"slug":8,"type":15},{"name":1492,"slug":1493,"type":15},"2026-07-17T05:29:03.913266",{"slug":1593,"name":1593,"fn":1594,"description":1595,"org":1596,"tags":1597,"stars":23,"repoUrl":24,"updatedAt":1606},"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},[1598,1599,1602,1603],{"name":1536,"slug":1537,"type":15},{"name":1600,"slug":1601,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":1604,"slug":1605,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":1608,"name":1608,"fn":1609,"description":1610,"org":1611,"tags":1612,"stars":23,"repoUrl":24,"updatedAt":1621},"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},[1613,1614,1617,1618],{"name":1410,"slug":1411,"type":15},{"name":1615,"slug":1616,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":1619,"slug":1620,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":1623,"name":1623,"fn":1624,"description":1625,"org":1626,"tags":1627,"stars":23,"repoUrl":24,"updatedAt":1635},"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},[1628,1629,1632],{"name":9,"slug":8,"type":15},{"name":1630,"slug":1631,"type":15},"Quantum Computing","quantum-computing",{"name":1633,"slug":1634,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305]