[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-physicsnemo-discover":3,"mdc-u37382-key":34,"related-repo-nvidia-physicsnemo-discover":538,"related-org-nvidia-physicsnemo-discover":642},{"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},"physicsnemo-discover","navigate PhysicsNeMo models and datapipe examples","Official NVIDIA-authored guidance for navigating PhysicsNeMo — pick the model, datapipe, or example for a SciML\u002FAI4Science task (surrogates, forecasting, downscaling, physics-informed, inverse, generative). Points at existing files via live repo search; never writes code. Do NOT use for installation or environment setup, training-loop or other code authoring\u002Fscaffolding, contributor\u002FCI\u002Fpackaging questions, repo-specific questions in physicsnemo-sym\u002F-cfd\u002F-curator, or general (non-physics) ML\u002FPyTorch.",{"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},"Research","research","tag",{"name":17,"slug":18,"type":15},"Machine Learning","machine-learning",{"name":9,"slug":8,"type":15},{"name":21,"slug":22,"type":15},"Physics","physics",2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-07-14T05:26:17.641182","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\u002Fphysicsnemo-discover","---\nname: physicsnemo-discover\ndescription: Official NVIDIA-authored guidance for navigating PhysicsNeMo — pick the model, datapipe, or example for a SciML\u002FAI4Science task (surrogates, forecasting, downscaling, physics-informed, inverse, generative). Points at existing files via live repo search; never writes code. Do NOT use for installation or environment setup, training-loop or other code authoring\u002Fscaffolding, contributor\u002FCI\u002Fpackaging questions, repo-specific questions in physicsnemo-sym\u002F-cfd\u002F-curator, or general (non-physics) ML\u002FPyTorch.\nlicense: Apache-2.0\nmetadata:\n  author: NVIDIA \u003Cagent-skills@nvidia.com>\n  tags:\n    - physicsnemo\n    - sciml\n    - ai4science\n    - discovery\n    - routing\n---\n\n# PhysicsNeMo Discoverability\n\nHelp a user navigate PhysicsNeMo: point them at files, folders, examples, and docs **in the repo at its current state**. Never write training code; never cite a path from memory.\n\n## Core principle\n\nPhysicsNeMo evolves — classes get renamed, examples move, `experimental\u002F` graduates. Any static list of class names and paths rots, so **discover, don't remember**: enumerate from the live repo every turn.\n\nPhysicsNeMo is **composable**: each solution is a product (model family × datapipe × training strategy × config). An example is one reference instantiation of that product, not a prescription. Surface the **axes** and the **menu along each axis**, then cite examples as concrete starting points to fork and recombine.\n\n## What a correct answer satisfies\n\nThese are constraints, not a script — choose the searches that meet them and skip work the task doesn't need. Search patterns per axis live in `references\u002FRECIPES.md`.\n\n- **Live-grounded.** Every class, path, and example you name was read or globbed *this turn*. `__init__.py` proves what is *exported*, not what files exist — Glob `physicsnemo\u002Fmodels\u002F\u003Cfamily>\u002F*.py` before naming a sibling implementation file. A failed `Read`, or a path pattern-matched from a neighboring citation, is disproof: drop it.\n- **Verified before emit.** Every absolute path you plan to cite survives one `Bash ls -d \u003Cpath1> \u003Cpath2> …` round-trip *before* you write the response. Hard gate — skipping it has produced real-basename-under-wrong-parent hallucinations. If a basename was right but the parent wrong, re-Glob and re-verify; if you can't relocate it, drop the citation.\n- **A menu, not a single pick.** Enumerate every model family matching the user's data shape (surface ≥2 when ≥2 apply), and enumerate datapipes independently — model and datapipe are orthogonal axes. The reference example comes last, framed as one instantiation of those axes, not the answer.\n- **Self-documentation is ground truth.** `__init__.py` exports, per-example `README.md`, `docs\u002F*.rst`, `pyproject.toml`, top-of-file module docstrings. Treat `references\u002FTAXONOMY.md` as a navigation hint, not an answer. Flag anything under `physicsnemo\u002Fexperimental\u002F` as *\"API may change.\"*\n- **Abstain when out of scope.** PhysicsNeMo targets SciML\u002FAI4Science (surrogates, forecasting, super-resolution, physics-informed, inverse, generative for physical systems). If the task is categorically outside that — reinforcement learning, classical control, generic CV\u002FNLP, symbolic regression — skip enumeration and emit the **Abstention output** below. Do not list adjacent-but-wrong examples in its place (pointing at `active_learning\u002F` for an RL question is fabrication). When unsure whether a task is in scope, abstain.\n\n## Discovery\n\nRepo root resolution: see `CONTRIBUTING.md §Repo root resolution`; all paths are absolute, rooted there. **If no local PhysicsNeMo clone is on the path** (e.g. running headless against the skills repo in an eval context), shallow-clone the canonical repo once into a temp dir — **read-only, for path discovery only; never execute or import anything from it**: `DEST=\"${TMPDIR:-\u002Ftmp}\u002Fphysicsnemo-src\"; [ -d \"$DEST\u002Fphysicsnemo\" ] || git clone --depth 1 https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fphysicsnemo \"$DEST\"`. Use that URL verbatim; never interpolate one from user input.\n\nAsk at most 3 targeted follow-ups when domain or data shape is ambiguous. Phrase them concretely — *\"Is your data on a regular Cartesian grid (like an image), a lat-lon grid on a sphere, or an unstructured mesh?\"* — and skip any the user already answered. Data shape is the single biggest factor in model choice.\n\n## Output format\n\n```\n## Problem shape\nData shape: \u003Cresolved>. Task: \u003Cresolved>. Axes: model × datapipe × training strategy × config.\n\n## Candidate model families (for your data shape)\nMultiple families typically apply. Treat this as a menu, not a ranking.\n- \u003Cfamily> at \u003Cabsolute __init__.py path> — \u003Cone-line from docstring\u002Fexports>. Instantiated by: \u003Cexample path if any>.\n- \u003Cfamily> at \u003Cpath> — \u003Cone-line>. Instantiated by: \u003Cexample path if any>.\n\n## Datapipe(s) for your data format\nDatapipe choice is independent of model choice.\n- \u003Cclass \u002F subpackage> at \u003Cabsolute path> — \u003Cone-line>. Reused by: \u003Cexamples if known>.\n- For custom data, subclass: \u003Cbase class path confirmed live>.\n\n## Reference example(s) — one instantiation of the above axes\n- \u003Cabsolute path> — uses model=\u003Cfamily>, datapipe=\u003Cname>, strategy=\u003Csingle-GPU|DDP|FSDP|...>.\n  Why it matches: \u003Cone line>.\n\n## Supporting docs\n- \u003Cabsolute path> — \u003Cone-line scope>\n\n## Suggested reading order\n1. \u003Cmodels\u002F\u003Cfamily>\u002F__init__.py> — survey alternative families\n2. \u003Cdatapipe __init__.py or base-class file> — understand the data axis\n3. \u003Cexample path> — concrete end-to-end instantiation to fork\n```\n\n**Rules for the output:**\n- Absolute paths only; every one survived the `ls -d` gate.\n- Every pointer needs a one-line justification grounded in content you actually read.\n- Caps: **4 model families** (minimum 2 when ≥2 exist), **3 datapipes**, **2 reference examples**, **2 docs**.\n- Name which (model, datapipe, strategy) axes each example fills.\n- If ≥2 model families apply, say so: *\"Other model families apply to the same data shape — see the candidate list above.\"*\n- End with the suggested reading order. Offer 2-3 forward steps (config file, training script, `experimental\u002F` look-alikes); do not start writing code unless asked.\n\n## Abstention output\n\nWhen out of scope, replace the menu skeleton with this shape — three sections, in this order, none skipped:\n\n```\n## PhysicsNeMo does not have direct support for \u003Cuser's problem class>\nOne sentence on why it's outside scope (e.g., \"PhysicsNeMo targets physics\nsurrogates and forecasting; reinforcement learning for molecular design is\nnot in its scope\").\n\n## Where to look instead\n- \u003Csibling NVIDIA framework or external library> at \u003CURL or repo name> — \u003Cone-line on why it fits>.\n- (One or two alternatives is enough; do not invent libraries.)\n\n## If you still want to build it in PhysicsNeMo\nConfirm the closest base classes by Reading `physicsnemo\u002Fcore\u002F__init__.py` and\n`physicsnemo\u002Fdatapipes\u002F__init__.py` first; then name them as subclassing\ntargets. This is the fallback, not the recommendation.\n```\n\n**Do not** open with the menu skeleton and bury \"no match\" at the end. **Do not** invent external libraries — if you don't know the right alternative, stop at the first two sections.\n\n## Related resources\n\n- `references\u002FTAXONOMY.md` — navigation hints (data-shape → folder mappings, decision axes, stability tiers).\n- `references\u002FRECIPES.md` — concrete Glob\u002FGrep\u002FRead patterns per discovery axis.\n",{"data":35,"body":44},{"name":4,"description":6,"license":26,"metadata":36},{"author":37,"tags":38},"NVIDIA \u003Cagent-skills@nvidia.com>",[39,40,41,42,43],"physicsnemo","sciml","ai4science","discovery","routing",{"type":45,"children":46},"root",[47,56,70,77,98,124,130,143,318,323,358,370,376,388,396,474,479,484,493,509,515],{"type":48,"tag":49,"props":50,"children":52},"element","h1",{"id":51},"physicsnemo-discoverability",[53],{"type":54,"value":55},"text","PhysicsNeMo Discoverability",{"type":48,"tag":57,"props":58,"children":59},"p",{},[60,62,68],{"type":54,"value":61},"Help a user navigate PhysicsNeMo: point them at files, folders, examples, and docs ",{"type":48,"tag":63,"props":64,"children":65},"strong",{},[66],{"type":54,"value":67},"in the repo at its current state",{"type":54,"value":69},". Never write training code; never cite a path from memory.",{"type":48,"tag":71,"props":72,"children":74},"h2",{"id":73},"core-principle",[75],{"type":54,"value":76},"Core principle",{"type":48,"tag":57,"props":78,"children":79},{},[80,82,89,91,96],{"type":54,"value":81},"PhysicsNeMo evolves — classes get renamed, examples move, ",{"type":48,"tag":83,"props":84,"children":86},"code",{"className":85},[],[87],{"type":54,"value":88},"experimental\u002F",{"type":54,"value":90}," graduates. Any static list of class names and paths rots, so ",{"type":48,"tag":63,"props":92,"children":93},{},[94],{"type":54,"value":95},"discover, don't remember",{"type":54,"value":97},": enumerate from the live repo every turn.",{"type":48,"tag":57,"props":99,"children":100},{},[101,103,108,110,115,117,122],{"type":54,"value":102},"PhysicsNeMo is ",{"type":48,"tag":63,"props":104,"children":105},{},[106],{"type":54,"value":107},"composable",{"type":54,"value":109},": each solution is a product (model family × datapipe × training strategy × config). An example is one reference instantiation of that product, not a prescription. Surface the ",{"type":48,"tag":63,"props":111,"children":112},{},[113],{"type":54,"value":114},"axes",{"type":54,"value":116}," and the ",{"type":48,"tag":63,"props":118,"children":119},{},[120],{"type":54,"value":121},"menu along each axis",{"type":54,"value":123},", then cite examples as concrete starting points to fork and recombine.",{"type":48,"tag":71,"props":125,"children":127},{"id":126},"what-a-correct-answer-satisfies",[128],{"type":54,"value":129},"What a correct answer satisfies",{"type":48,"tag":57,"props":131,"children":132},{},[133,135,141],{"type":54,"value":134},"These are constraints, not a script — choose the searches that meet them and skip work the task doesn't need. Search patterns per axis live in ",{"type":48,"tag":83,"props":136,"children":138},{"className":137},[],[139],{"type":54,"value":140},"references\u002FRECIPES.md",{"type":54,"value":142},".",{"type":48,"tag":144,"props":145,"children":146},"ul",{},[147,197,222,232,293],{"type":48,"tag":148,"props":149,"children":150},"li",{},[151,156,158,164,166,172,174,179,181,187,189,195],{"type":48,"tag":63,"props":152,"children":153},{},[154],{"type":54,"value":155},"Live-grounded.",{"type":54,"value":157}," Every class, path, and example you name was read or globbed ",{"type":48,"tag":159,"props":160,"children":161},"em",{},[162],{"type":54,"value":163},"this turn",{"type":54,"value":165},". ",{"type":48,"tag":83,"props":167,"children":169},{"className":168},[],[170],{"type":54,"value":171},"__init__.py",{"type":54,"value":173}," proves what is ",{"type":48,"tag":159,"props":175,"children":176},{},[177],{"type":54,"value":178},"exported",{"type":54,"value":180},", not what files exist — Glob ",{"type":48,"tag":83,"props":182,"children":184},{"className":183},[],[185],{"type":54,"value":186},"physicsnemo\u002Fmodels\u002F\u003Cfamily>\u002F*.py",{"type":54,"value":188}," before naming a sibling implementation file. A failed ",{"type":48,"tag":83,"props":190,"children":192},{"className":191},[],[193],{"type":54,"value":194},"Read",{"type":54,"value":196},", or a path pattern-matched from a neighboring citation, is disproof: drop it.",{"type":48,"tag":148,"props":198,"children":199},{},[200,205,207,213,215,220],{"type":48,"tag":63,"props":201,"children":202},{},[203],{"type":54,"value":204},"Verified before emit.",{"type":54,"value":206}," Every absolute path you plan to cite survives one ",{"type":48,"tag":83,"props":208,"children":210},{"className":209},[],[211],{"type":54,"value":212},"Bash ls -d \u003Cpath1> \u003Cpath2> …",{"type":54,"value":214}," round-trip ",{"type":48,"tag":159,"props":216,"children":217},{},[218],{"type":54,"value":219},"before",{"type":54,"value":221}," you write the response. Hard gate — skipping it has produced real-basename-under-wrong-parent hallucinations. If a basename was right but the parent wrong, re-Glob and re-verify; if you can't relocate it, drop the citation.",{"type":48,"tag":148,"props":223,"children":224},{},[225,230],{"type":48,"tag":63,"props":226,"children":227},{},[228],{"type":54,"value":229},"A menu, not a single pick.",{"type":54,"value":231}," Enumerate every model family matching the user's data shape (surface ≥2 when ≥2 apply), and enumerate datapipes independently — model and datapipe are orthogonal axes. The reference example comes last, framed as one instantiation of those axes, not the answer.",{"type":48,"tag":148,"props":233,"children":234},{},[235,240,242,247,249,255,257,263,264,270,272,278,280,286,288],{"type":48,"tag":63,"props":236,"children":237},{},[238],{"type":54,"value":239},"Self-documentation is ground truth.",{"type":54,"value":241}," ",{"type":48,"tag":83,"props":243,"children":245},{"className":244},[],[246],{"type":54,"value":171},{"type":54,"value":248}," exports, per-example ",{"type":48,"tag":83,"props":250,"children":252},{"className":251},[],[253],{"type":54,"value":254},"README.md",{"type":54,"value":256},", ",{"type":48,"tag":83,"props":258,"children":260},{"className":259},[],[261],{"type":54,"value":262},"docs\u002F*.rst",{"type":54,"value":256},{"type":48,"tag":83,"props":265,"children":267},{"className":266},[],[268],{"type":54,"value":269},"pyproject.toml",{"type":54,"value":271},", top-of-file module docstrings. Treat ",{"type":48,"tag":83,"props":273,"children":275},{"className":274},[],[276],{"type":54,"value":277},"references\u002FTAXONOMY.md",{"type":54,"value":279}," as a navigation hint, not an answer. Flag anything under ",{"type":48,"tag":83,"props":281,"children":283},{"className":282},[],[284],{"type":54,"value":285},"physicsnemo\u002Fexperimental\u002F",{"type":54,"value":287}," as ",{"type":48,"tag":159,"props":289,"children":290},{},[291],{"type":54,"value":292},"\"API may change.\"",{"type":48,"tag":148,"props":294,"children":295},{},[296,301,303,308,310,316],{"type":48,"tag":63,"props":297,"children":298},{},[299],{"type":54,"value":300},"Abstain when out of scope.",{"type":54,"value":302}," PhysicsNeMo targets SciML\u002FAI4Science (surrogates, forecasting, super-resolution, physics-informed, inverse, generative for physical systems). If the task is categorically outside that — reinforcement learning, classical control, generic CV\u002FNLP, symbolic regression — skip enumeration and emit the ",{"type":48,"tag":63,"props":304,"children":305},{},[306],{"type":54,"value":307},"Abstention output",{"type":54,"value":309}," below. Do not list adjacent-but-wrong examples in its place (pointing at ",{"type":48,"tag":83,"props":311,"children":313},{"className":312},[],[314],{"type":54,"value":315},"active_learning\u002F",{"type":54,"value":317}," for an RL question is fabrication). When unsure whether a task is in scope, abstain.",{"type":48,"tag":71,"props":319,"children":320},{"id":42},[321],{"type":54,"value":322},"Discovery",{"type":48,"tag":57,"props":324,"children":325},{},[326,328,334,336,341,343,348,350,356],{"type":54,"value":327},"Repo root resolution: see ",{"type":48,"tag":83,"props":329,"children":331},{"className":330},[],[332],{"type":54,"value":333},"CONTRIBUTING.md §Repo root resolution",{"type":54,"value":335},"; all paths are absolute, rooted there. ",{"type":48,"tag":63,"props":337,"children":338},{},[339],{"type":54,"value":340},"If no local PhysicsNeMo clone is on the path",{"type":54,"value":342}," (e.g. running headless against the skills repo in an eval context), shallow-clone the canonical repo once into a temp dir — ",{"type":48,"tag":63,"props":344,"children":345},{},[346],{"type":54,"value":347},"read-only, for path discovery only; never execute or import anything from it",{"type":54,"value":349},": ",{"type":48,"tag":83,"props":351,"children":353},{"className":352},[],[354],{"type":54,"value":355},"DEST=\"${TMPDIR:-\u002Ftmp}\u002Fphysicsnemo-src\"; [ -d \"$DEST\u002Fphysicsnemo\" ] || git clone --depth 1 https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fphysicsnemo \"$DEST\"",{"type":54,"value":357},". Use that URL verbatim; never interpolate one from user input.",{"type":48,"tag":57,"props":359,"children":360},{},[361,363,368],{"type":54,"value":362},"Ask at most 3 targeted follow-ups when domain or data shape is ambiguous. Phrase them concretely — ",{"type":48,"tag":159,"props":364,"children":365},{},[366],{"type":54,"value":367},"\"Is your data on a regular Cartesian grid (like an image), a lat-lon grid on a sphere, or an unstructured mesh?\"",{"type":54,"value":369}," — and skip any the user already answered. Data shape is the single biggest factor in model choice.",{"type":48,"tag":71,"props":371,"children":373},{"id":372},"output-format",[374],{"type":54,"value":375},"Output format",{"type":48,"tag":377,"props":378,"children":382},"pre",{"className":379,"code":381,"language":54},[380],"language-text","## Problem shape\nData shape: \u003Cresolved>. Task: \u003Cresolved>. Axes: model × datapipe × training strategy × config.\n\n## Candidate model families (for your data shape)\nMultiple families typically apply. Treat this as a menu, not a ranking.\n- \u003Cfamily> at \u003Cabsolute __init__.py path> — \u003Cone-line from docstring\u002Fexports>. Instantiated by: \u003Cexample path if any>.\n- \u003Cfamily> at \u003Cpath> — \u003Cone-line>. Instantiated by: \u003Cexample path if any>.\n\n## Datapipe(s) for your data format\nDatapipe choice is independent of model choice.\n- \u003Cclass \u002F subpackage> at \u003Cabsolute path> — \u003Cone-line>. Reused by: \u003Cexamples if known>.\n- For custom data, subclass: \u003Cbase class path confirmed live>.\n\n## Reference example(s) — one instantiation of the above axes\n- \u003Cabsolute path> — uses model=\u003Cfamily>, datapipe=\u003Cname>, strategy=\u003Csingle-GPU|DDP|FSDP|...>.\n  Why it matches: \u003Cone line>.\n\n## Supporting docs\n- \u003Cabsolute path> — \u003Cone-line scope>\n\n## Suggested reading order\n1. \u003Cmodels\u002F\u003Cfamily>\u002F__init__.py> — survey alternative families\n2. \u003Cdatapipe __init__.py or base-class file> — understand the data axis\n3. \u003Cexample path> — concrete end-to-end instantiation to fork\n",[383],{"type":48,"tag":83,"props":384,"children":386},{"__ignoreMap":385},"",[387],{"type":54,"value":381},{"type":48,"tag":57,"props":389,"children":390},{},[391],{"type":48,"tag":63,"props":392,"children":393},{},[394],{"type":54,"value":395},"Rules for the output:",{"type":48,"tag":144,"props":397,"children":398},{},[399,412,417,447,452,462],{"type":48,"tag":148,"props":400,"children":401},{},[402,404,410],{"type":54,"value":403},"Absolute paths only; every one survived the ",{"type":48,"tag":83,"props":405,"children":407},{"className":406},[],[408],{"type":54,"value":409},"ls -d",{"type":54,"value":411}," gate.",{"type":48,"tag":148,"props":413,"children":414},{},[415],{"type":54,"value":416},"Every pointer needs a one-line justification grounded in content you actually read.",{"type":48,"tag":148,"props":418,"children":419},{},[420,422,427,429,434,435,440,441,446],{"type":54,"value":421},"Caps: ",{"type":48,"tag":63,"props":423,"children":424},{},[425],{"type":54,"value":426},"4 model families",{"type":54,"value":428}," (minimum 2 when ≥2 exist), ",{"type":48,"tag":63,"props":430,"children":431},{},[432],{"type":54,"value":433},"3 datapipes",{"type":54,"value":256},{"type":48,"tag":63,"props":436,"children":437},{},[438],{"type":54,"value":439},"2 reference examples",{"type":54,"value":256},{"type":48,"tag":63,"props":442,"children":443},{},[444],{"type":54,"value":445},"2 docs",{"type":54,"value":142},{"type":48,"tag":148,"props":448,"children":449},{},[450],{"type":54,"value":451},"Name which (model, datapipe, strategy) axes each example fills.",{"type":48,"tag":148,"props":453,"children":454},{},[455,457],{"type":54,"value":456},"If ≥2 model families apply, say so: ",{"type":48,"tag":159,"props":458,"children":459},{},[460],{"type":54,"value":461},"\"Other model families apply to the same data shape — see the candidate list above.\"",{"type":48,"tag":148,"props":463,"children":464},{},[465,467,472],{"type":54,"value":466},"End with the suggested reading order. Offer 2-3 forward steps (config file, training script, ",{"type":48,"tag":83,"props":468,"children":470},{"className":469},[],[471],{"type":54,"value":88},{"type":54,"value":473}," look-alikes); do not start writing code unless asked.",{"type":48,"tag":71,"props":475,"children":477},{"id":476},"abstention-output",[478],{"type":54,"value":307},{"type":48,"tag":57,"props":480,"children":481},{},[482],{"type":54,"value":483},"When out of scope, replace the menu skeleton with this shape — three sections, in this order, none skipped:",{"type":48,"tag":377,"props":485,"children":488},{"className":486,"code":487,"language":54},[380],"## PhysicsNeMo does not have direct support for \u003Cuser's problem class>\nOne sentence on why it's outside scope (e.g., \"PhysicsNeMo targets physics\nsurrogates and forecasting; reinforcement learning for molecular design is\nnot in its scope\").\n\n## Where to look instead\n- \u003Csibling NVIDIA framework or external library> at \u003CURL or repo name> — \u003Cone-line on why it fits>.\n- (One or two alternatives is enough; do not invent libraries.)\n\n## If you still want to build it in PhysicsNeMo\nConfirm the closest base classes by Reading `physicsnemo\u002Fcore\u002F__init__.py` and\n`physicsnemo\u002Fdatapipes\u002F__init__.py` first; then name them as subclassing\ntargets. This is the fallback, not the recommendation.\n",[489],{"type":48,"tag":83,"props":490,"children":491},{"__ignoreMap":385},[492],{"type":54,"value":487},{"type":48,"tag":57,"props":494,"children":495},{},[496,501,503,507],{"type":48,"tag":63,"props":497,"children":498},{},[499],{"type":54,"value":500},"Do not",{"type":54,"value":502}," open with the menu skeleton and bury \"no match\" at the end. ",{"type":48,"tag":63,"props":504,"children":505},{},[506],{"type":54,"value":500},{"type":54,"value":508}," invent external libraries — if you don't know the right alternative, stop at the first two sections.",{"type":48,"tag":71,"props":510,"children":512},{"id":511},"related-resources",[513],{"type":54,"value":514},"Related resources",{"type":48,"tag":144,"props":516,"children":517},{},[518,528],{"type":48,"tag":148,"props":519,"children":520},{},[521,526],{"type":48,"tag":83,"props":522,"children":524},{"className":523},[],[525],{"type":54,"value":277},{"type":54,"value":527}," — navigation hints (data-shape → folder mappings, decision axes, stability tiers).",{"type":48,"tag":148,"props":529,"children":530},{},[531,536],{"type":48,"tag":83,"props":532,"children":534},{"className":533},[],[535],{"type":54,"value":140},{"type":54,"value":537}," — concrete Glob\u002FGrep\u002FRead patterns per discovery axis.",{"items":539,"total":641},[540,557,571,583,595,612,627],{"slug":541,"name":541,"fn":542,"description":543,"org":544,"tags":545,"stars":23,"repoUrl":24,"updatedAt":556},"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},[546,549,552,553],{"name":547,"slug":548,"type":15},"Data Analysis","data-analysis",{"name":550,"slug":551,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":554,"slug":555,"type":15},"Performance","performance","2026-07-14T05:28:43.176466",{"slug":558,"name":558,"fn":559,"description":560,"org":561,"tags":562,"stars":23,"repoUrl":24,"updatedAt":570},"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},[563,566,569],{"name":564,"slug":565,"type":15},"Deployment","deployment",{"name":567,"slug":568,"type":15},"Infrastructure","infrastructure",{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":572,"name":572,"fn":573,"description":574,"org":575,"tags":576,"stars":23,"repoUrl":24,"updatedAt":582},"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},[577,580,581],{"name":578,"slug":579,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":13,"slug":14,"type":15},"2026-07-14T05:28:06.816956",{"slug":584,"name":584,"fn":585,"description":586,"org":587,"tags":588,"stars":23,"repoUrl":24,"updatedAt":594},"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},[589,590,591],{"name":547,"slug":548,"type":15},{"name":9,"slug":8,"type":15},{"name":592,"slug":593,"type":15},"Testing","testing","2026-07-17T05:29:03.913266",{"slug":596,"name":596,"fn":597,"description":598,"org":599,"tags":600,"stars":23,"repoUrl":24,"updatedAt":611},"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},[601,604,607,608],{"name":602,"slug":603,"type":15},"Automation","automation",{"name":605,"slug":606,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":609,"slug":610,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":613,"name":613,"fn":614,"description":615,"org":616,"tags":617,"stars":23,"repoUrl":24,"updatedAt":626},"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},[618,619,622,623],{"name":564,"slug":565,"type":15},{"name":620,"slug":621,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":624,"slug":625,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":628,"name":628,"fn":629,"description":630,"org":631,"tags":632,"stars":23,"repoUrl":24,"updatedAt":640},"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},[633,634,637],{"name":9,"slug":8,"type":15},{"name":635,"slug":636,"type":15},"Quantum Computing","quantum-computing",{"name":638,"slug":639,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305,{"items":643,"total":792},[644,662,678,689,701,715,728,740,751,760,774,783],{"slug":645,"name":645,"fn":646,"description":647,"org":648,"tags":649,"stars":659,"repoUrl":660,"updatedAt":661},"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},[650,653,656],{"name":651,"slug":652,"type":15},"Documentation","documentation",{"name":654,"slug":655,"type":15},"MCP","mcp",{"name":657,"slug":658,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":663,"name":663,"fn":664,"description":665,"org":666,"tags":667,"stars":675,"repoUrl":676,"updatedAt":677},"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},[668,671,672],{"name":669,"slug":670,"type":15},"Containers","containers",{"name":564,"slug":565,"type":15},{"name":673,"slug":674,"type":15},"Python","python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":679,"name":679,"fn":680,"description":681,"org":682,"tags":683,"stars":675,"repoUrl":676,"updatedAt":688},"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},[684,687],{"name":685,"slug":686,"type":15},"CI\u002FCD","ci-cd",{"name":564,"slug":565,"type":15},"2026-07-14T05:25:59.97109",{"slug":690,"name":690,"fn":691,"description":692,"org":693,"tags":694,"stars":675,"repoUrl":676,"updatedAt":700},"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},[695,696,697],{"name":685,"slug":686,"type":15},{"name":564,"slug":565,"type":15},{"name":698,"slug":699,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":702,"name":702,"fn":703,"description":704,"org":705,"tags":706,"stars":675,"repoUrl":676,"updatedAt":714},"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},[707,710,711],{"name":708,"slug":709,"type":15},"Debugging","debugging",{"name":698,"slug":699,"type":15},{"name":712,"slug":713,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":716,"name":716,"fn":717,"description":718,"org":719,"tags":720,"stars":675,"repoUrl":676,"updatedAt":727},"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},[721,724],{"name":722,"slug":723,"type":15},"Best Practices","best-practices",{"name":725,"slug":726,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":729,"name":729,"fn":730,"description":731,"org":732,"tags":733,"stars":675,"repoUrl":676,"updatedAt":739},"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},[734,735,738],{"name":17,"slug":18,"type":15},{"name":736,"slug":737,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":741,"name":741,"fn":742,"description":743,"org":744,"tags":745,"stars":675,"repoUrl":676,"updatedAt":750},"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},[746,749],{"name":747,"slug":748,"type":15},"QA","qa",{"name":592,"slug":593,"type":15},"2026-07-14T05:25:53.673039",{"slug":752,"name":752,"fn":753,"description":754,"org":755,"tags":756,"stars":675,"repoUrl":676,"updatedAt":759},"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},[757,758],{"name":564,"slug":565,"type":15},{"name":567,"slug":568,"type":15},"2026-07-14T05:25:49.362534",{"slug":761,"name":761,"fn":762,"description":763,"org":764,"tags":765,"stars":675,"repoUrl":676,"updatedAt":773},"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},[766,769,770],{"name":767,"slug":768,"type":15},"Code Review","code-review",{"name":698,"slug":699,"type":15},{"name":771,"slug":772,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":775,"name":775,"fn":776,"description":777,"org":778,"tags":779,"stars":675,"repoUrl":676,"updatedAt":782},"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},[780,781],{"name":747,"slug":748,"type":15},{"name":592,"slug":593,"type":15},"2026-07-14T05:25:54.928983",{"slug":784,"name":784,"fn":785,"description":786,"org":787,"tags":788,"stars":675,"repoUrl":676,"updatedAt":791},"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},[789,790],{"name":602,"slug":603,"type":15},{"name":685,"slug":686,"type":15},"2026-07-30T05:29:03.275638",496]