[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-nvshmem-tune-performance":3,"mdc--db3crl-key":36,"related-repo-nvidia-nvshmem-tune-performance":607,"related-org-nvidia-nvshmem-tune-performance":694},{"slug":4,"name":4,"fn":5,"description":6,"org":7,"tags":11,"stars":20,"repoUrl":21,"updatedAt":22,"license":23,"forks":24,"topics":25,"repo":31,"sourceUrl":34,"mdContent":35},"nvshmem-tune-performance","tune NVSHMEM performance","Route NVSHMEM tuning to data collection, remote transport, NIC-to-PE mapping, or TMA. Do not use for unrelated CUDA, NCCL, or application tuning.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},"nvidia","NVIDIA","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fnvidia.png",[12,16,17],{"name":13,"slug":14,"type":15},"Performance","performance","tag",{"name":9,"slug":8,"type":15},{"name":18,"slug":19,"type":15},"Engineering","engineering",579,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fnvshmem","2026-08-31T09:18:56.180659","Apache-2.0",104,[26,27,28,29,8,30],"communciations","cpp","cuda","deep-learning","python",{"repoUrl":21,"stars":20,"forks":24,"topics":32,"description":33},[26,27,28,29,8,30],"NVIDIA NVSHMEM is a parallel programming interface for NVIDIA GPUs based on OpenSHMEM. NVSHMEM can significantly reduce multi-process communication and coordination overheads by allowing programmers to perform one-sided communication from within CUDA kernels and on CUDA streams.","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fnvshmem\u002Ftree\u002FHEAD\u002Fskills\u002Fnvshmem-tune-performance","---\nname: nvshmem-tune-performance\ndescription: Route NVSHMEM tuning to data collection, remote transport, NIC-to-PE mapping, or TMA. Do not use for unrelated CUDA, NCCL, or application tuning.\nlicense: Apache-2.0\nmetadata:\n  version: \"1.0.0\"\n  author: NVIDIA NVSHMEM Team \u003Cnvshmem@nvidia.com>\n  tags:\n    - nvshmem\n    - performance\n    - tuning\n---\n\n# Tune NVSHMEM Performance\n\n## Purpose\n\nGuide the user to the smallest applicable NVSHMEM performance workflow. Offer optional baseline collection and three focused optimization services while preserving each specialist's inputs, execution boundary, and safety rules.\n\nDo not reproduce the specialists' technical guidance in this router. Invoke:\n\n| User goal | Specialist |\n| --- | --- |\n| Collect a baseline, run perftests, or package system and benchmark evidence | `$nvshmem-collect-performance-data` |\n| Select an inter-node remote transport | `$nvshmem-select-remote-transport` |\n| Select HCAs or optimize NIC-to-PE mapping | `$nvshmem-configure-nic-pe-mapping` |\n| Assess, prepare, review, or debug NVSHMEM TMA use | `$nvshmem-enable-tma` |\n\n## Requirements\n\n- Make the four specialist skills listed above available; this router delegates all technical work to them.\n- Require no API keys, cluster access, NVSHMEM installation, or benchmark tools to select a route. The selected specialist establishes any environment-specific prerequisites before acting.\n\n## Inputs\n\n### Required\n\n- Obtain the user's NVSHMEM performance goal or selected service. When the request is vague, obtain the user's choice from the menu in the instructions before invoking a specialist.\n\n### Optional\n\n- Reuse any supplied performance report, raw logs, exact NVSHMEM version and installation prefix, source or kernel path, fabric and provider, job scale, PE-to-GPU binding, GPU\u002FNIC topology, selected transport, tuning constraints, and optimization goal.\n- Reuse relevant results and unresolved conditions from an earlier specialist in the same workflow.\n- Ask only for information needed to choose the route. Let the selected specialist inspect the environment or request its own required inputs.\n\nPrefer the current explicit user request, then explicit invocation arguments, then non-conflicting context and prior specialist output. Report conflicting values instead of choosing silently.\n\n## Instructions\n\n### 1. Choose a Route\n\nHonor an explicitly named specialist or workflow. Otherwise classify the concrete objective:\n\n- Route requests for performance baselines, benchmark curves, latency or bandwidth perftests, performance sanity checks, or packaged system evidence to performance data collection.\n- Route requests about `NVSHMEM_REMOTE_TRANSPORT`, IBRC, IBDevX, IBGDA, GPUNetIO, UCX, libfabric, remote-network compatibility, or choosing an inter-node data path to remote transport selection.\n- Route requests about HCA selection, NIC ports, GPU-to-NIC affinity, multi-NIC use, `NVSHMEM_HCA_*`, or PE-to-NIC assignment to NIC-to-PE mapping.\n- Route requests about TMA, `NVSHMEM_TMA_POLICY`, CTA shared-memory registration, direct shared-memory transfers, or preparing a CUDA kernel for TMA to TMA enablement.\n\nInvoke one specialist for a single concrete request. State which specialist is being invoked and why.\n\n### 2. Present the Menu for a Vague Request\n\nWhen the user asks only to improve or tune NVSHMEM performance, do not choose a specialist or run anything. Present these paths and ask the user to select one:\n\n1. **Baseline first:** collect initial performance and system data, then choose an optimization workflow.\n2. **Optimize directly:** skip collection and choose one of:\n   - remote transport selection for inter-node communication;\n   - NIC-to-PE mapping for topology-aware remote-transport placement;\n   - TMA assessment or enablement for eligible peer GPU\u002FNVLink kernel paths.\n\nAllow the user to collect data only and stop. Briefly distinguish the three optimization choices without recommending one from absent evidence.\n\n### 3. Sequence Explicit Multi-Stage Workflows\n\n1. Invoke performance data collection first only when the user requests it, selects it from the menu, or explicitly chooses a baseline-first sequence. Let that skill obtain its required benchmark selection before any run.\n2. Invoke remote transport selection before NIC-to-PE mapping when both are requested. Pass the selected transport, version, provider, topology evidence, and unresolved conditions into the mapping workflow.\n3. Invoke TMA separately. Do not treat TMA as a remote-transport optimization or infer that inter-node benchmark evidence establishes TMA eligibility.\n4. Invoke no unrequested specialist. Let the active specialist resolve its own prerequisites instead of filling them in.\n\nAfter data collection completes, invoke an optimizer already selected by the user. If none was selected, present the three optimization choices again and wait.\n\n### 4. Preserve the Handoff\n\n- Pass the original request, explicit constraints, exact paths, commands, logs, configuration values, source scope, raw artifacts, resolved facts, and relevant prior specialist output.\n- Preserve the user's transport settings and tuning constraints during collection and later comparison.\n- Return the specialist's result without replacing its confidence, unknowns, validation status, or requested follow-up.\n- Keep benchmark execution, system inspection, source edits, and configuration changes inside the selected specialist's authorization and guardrails.\n\n### 5. Handle an Unavailable Specialist\n\nIf a required skill is unavailable, name it, explain which requested service cannot proceed, and ask the user to make it available or choose another service. Do not imitate or reconstruct the missing skill.\n\n## Limitations\n\n- Do not run benchmarks, collectors, launchers, or target applications directly from this router.\n- Do not edit CUDA source, generate transport exports, or synthesize NIC mappings directly from this router.\n- Do not make data collection an implicit prerequisite merely because an optimization might benefit from later measurement.\n- Do not promise a performance improvement or declare a regression without the evidence required by the selected specialist.\n- Do not treat TMA as useful for remote-network traffic; route it only as a separate peer GPU\u002FNVLink optimization workflow.\n- Do not route unrelated NCCL, CUDA-only, operating-system, or general application-performance questions to these NVSHMEM skills.\n\n## Examples\n\nRoute a request to collect evidence before choosing an optimization:\n\n```text\nUser: Collect a baseline for our two-node NVSHMEM all-to-all before tuning it.\nAction: Invoke $nvshmem-collect-performance-data and preserve the two-node and all-to-all constraints.\n```\n\nRoute an explicit network-configuration request directly:\n\n```text\nUser: Which NVSHMEM remote transport should we use with this InfiniBand cluster?\nAction: Invoke $nvshmem-select-remote-transport because the request concerns an inter-node data path.\n```\n\n## Troubleshooting\n\n| Error or condition | Cause | Solution |\n| --- | --- | --- |\n| No route can be selected | The request says only to improve performance. | Present the baseline-first and three direct-optimization options; wait for the user to choose. |\n| A requested specialist is unavailable | Its skill is not installed or enabled. | Name the unavailable specialist and ask the user to enable it or choose another route. |\n| Supplied settings conflict | The request and earlier context specify different topology, transport, or constraints. | Report the conflict and ask which value to preserve before handing off. |\n",{"data":37,"body":44},{"name":4,"description":6,"license":23,"metadata":38},{"version":39,"author":40,"tags":41},"1.0.0","NVIDIA NVSHMEM Team \u003Cnvshmem@nvidia.com>",[42,14,43],"nvshmem","tuning",{"type":45,"children":46},"root",[47,56,63,69,74,171,177,192,198,205,213,219,237,242,248,254,259,306,311,317,322,365,370,376,399,404,410,433,439,444,450,483,489,494,506,511,520,526],{"type":48,"tag":49,"props":50,"children":52},"element","h1",{"id":51},"tune-nvshmem-performance",[53],{"type":54,"value":55},"text","Tune NVSHMEM Performance",{"type":48,"tag":57,"props":58,"children":60},"h2",{"id":59},"purpose",[61],{"type":54,"value":62},"Purpose",{"type":48,"tag":64,"props":65,"children":66},"p",{},[67],{"type":54,"value":68},"Guide the user to the smallest applicable NVSHMEM performance workflow. Offer optional baseline collection and three focused optimization services while preserving each specialist's inputs, execution boundary, and safety rules.",{"type":48,"tag":64,"props":70,"children":71},{},[72],{"type":54,"value":73},"Do not reproduce the specialists' technical guidance in this router. Invoke:",{"type":48,"tag":75,"props":76,"children":77},"table",{},[78,97],{"type":48,"tag":79,"props":80,"children":81},"thead",{},[82],{"type":48,"tag":83,"props":84,"children":85},"tr",{},[86,92],{"type":48,"tag":87,"props":88,"children":89},"th",{},[90],{"type":54,"value":91},"User goal",{"type":48,"tag":87,"props":93,"children":94},{},[95],{"type":54,"value":96},"Specialist",{"type":48,"tag":98,"props":99,"children":100},"tbody",{},[101,120,137,154],{"type":48,"tag":83,"props":102,"children":103},{},[104,110],{"type":48,"tag":105,"props":106,"children":107},"td",{},[108],{"type":54,"value":109},"Collect a baseline, run perftests, or package system and benchmark evidence",{"type":48,"tag":105,"props":111,"children":112},{},[113],{"type":48,"tag":114,"props":115,"children":117},"code",{"className":116},[],[118],{"type":54,"value":119},"$nvshmem-collect-performance-data",{"type":48,"tag":83,"props":121,"children":122},{},[123,128],{"type":48,"tag":105,"props":124,"children":125},{},[126],{"type":54,"value":127},"Select an inter-node remote transport",{"type":48,"tag":105,"props":129,"children":130},{},[131],{"type":48,"tag":114,"props":132,"children":134},{"className":133},[],[135],{"type":54,"value":136},"$nvshmem-select-remote-transport",{"type":48,"tag":83,"props":138,"children":139},{},[140,145],{"type":48,"tag":105,"props":141,"children":142},{},[143],{"type":54,"value":144},"Select HCAs or optimize NIC-to-PE mapping",{"type":48,"tag":105,"props":146,"children":147},{},[148],{"type":48,"tag":114,"props":149,"children":151},{"className":150},[],[152],{"type":54,"value":153},"$nvshmem-configure-nic-pe-mapping",{"type":48,"tag":83,"props":155,"children":156},{},[157,162],{"type":48,"tag":105,"props":158,"children":159},{},[160],{"type":54,"value":161},"Assess, prepare, review, or debug NVSHMEM TMA use",{"type":48,"tag":105,"props":163,"children":164},{},[165],{"type":48,"tag":114,"props":166,"children":168},{"className":167},[],[169],{"type":54,"value":170},"$nvshmem-enable-tma",{"type":48,"tag":57,"props":172,"children":174},{"id":173},"requirements",[175],{"type":54,"value":176},"Requirements",{"type":48,"tag":178,"props":179,"children":180},"ul",{},[181,187],{"type":48,"tag":182,"props":183,"children":184},"li",{},[185],{"type":54,"value":186},"Make the four specialist skills listed above available; this router delegates all technical work to them.",{"type":48,"tag":182,"props":188,"children":189},{},[190],{"type":54,"value":191},"Require no API keys, cluster access, NVSHMEM installation, or benchmark tools to select a route. The selected specialist establishes any environment-specific prerequisites before acting.",{"type":48,"tag":57,"props":193,"children":195},{"id":194},"inputs",[196],{"type":54,"value":197},"Inputs",{"type":48,"tag":199,"props":200,"children":202},"h3",{"id":201},"required",[203],{"type":54,"value":204},"Required",{"type":48,"tag":178,"props":206,"children":207},{},[208],{"type":48,"tag":182,"props":209,"children":210},{},[211],{"type":54,"value":212},"Obtain the user's NVSHMEM performance goal or selected service. When the request is vague, obtain the user's choice from the menu in the instructions before invoking a specialist.",{"type":48,"tag":199,"props":214,"children":216},{"id":215},"optional",[217],{"type":54,"value":218},"Optional",{"type":48,"tag":178,"props":220,"children":221},{},[222,227,232],{"type":48,"tag":182,"props":223,"children":224},{},[225],{"type":54,"value":226},"Reuse any supplied performance report, raw logs, exact NVSHMEM version and installation prefix, source or kernel path, fabric and provider, job scale, PE-to-GPU binding, GPU\u002FNIC topology, selected transport, tuning constraints, and optimization goal.",{"type":48,"tag":182,"props":228,"children":229},{},[230],{"type":54,"value":231},"Reuse relevant results and unresolved conditions from an earlier specialist in the same workflow.",{"type":48,"tag":182,"props":233,"children":234},{},[235],{"type":54,"value":236},"Ask only for information needed to choose the route. Let the selected specialist inspect the environment or request its own required inputs.",{"type":48,"tag":64,"props":238,"children":239},{},[240],{"type":54,"value":241},"Prefer the current explicit user request, then explicit invocation arguments, then non-conflicting context and prior specialist output. Report conflicting values instead of choosing silently.",{"type":48,"tag":57,"props":243,"children":245},{"id":244},"instructions",[246],{"type":54,"value":247},"Instructions",{"type":48,"tag":199,"props":249,"children":251},{"id":250},"_1-choose-a-route",[252],{"type":54,"value":253},"1. Choose a Route",{"type":48,"tag":64,"props":255,"children":256},{},[257],{"type":54,"value":258},"Honor an explicitly named specialist or workflow. Otherwise classify the concrete objective:",{"type":48,"tag":178,"props":260,"children":261},{},[262,267,280,293],{"type":48,"tag":182,"props":263,"children":264},{},[265],{"type":54,"value":266},"Route requests for performance baselines, benchmark curves, latency or bandwidth perftests, performance sanity checks, or packaged system evidence to performance data collection.",{"type":48,"tag":182,"props":268,"children":269},{},[270,272,278],{"type":54,"value":271},"Route requests about ",{"type":48,"tag":114,"props":273,"children":275},{"className":274},[],[276],{"type":54,"value":277},"NVSHMEM_REMOTE_TRANSPORT",{"type":54,"value":279},", IBRC, IBDevX, IBGDA, GPUNetIO, UCX, libfabric, remote-network compatibility, or choosing an inter-node data path to remote transport selection.",{"type":48,"tag":182,"props":281,"children":282},{},[283,285,291],{"type":54,"value":284},"Route requests about HCA selection, NIC ports, GPU-to-NIC affinity, multi-NIC use, ",{"type":48,"tag":114,"props":286,"children":288},{"className":287},[],[289],{"type":54,"value":290},"NVSHMEM_HCA_*",{"type":54,"value":292},", or PE-to-NIC assignment to NIC-to-PE mapping.",{"type":48,"tag":182,"props":294,"children":295},{},[296,298,304],{"type":54,"value":297},"Route requests about TMA, ",{"type":48,"tag":114,"props":299,"children":301},{"className":300},[],[302],{"type":54,"value":303},"NVSHMEM_TMA_POLICY",{"type":54,"value":305},", CTA shared-memory registration, direct shared-memory transfers, or preparing a CUDA kernel for TMA to TMA enablement.",{"type":48,"tag":64,"props":307,"children":308},{},[309],{"type":54,"value":310},"Invoke one specialist for a single concrete request. State which specialist is being invoked and why.",{"type":48,"tag":199,"props":312,"children":314},{"id":313},"_2-present-the-menu-for-a-vague-request",[315],{"type":54,"value":316},"2. Present the Menu for a Vague Request",{"type":48,"tag":64,"props":318,"children":319},{},[320],{"type":54,"value":321},"When the user asks only to improve or tune NVSHMEM performance, do not choose a specialist or run anything. Present these paths and ask the user to select one:",{"type":48,"tag":323,"props":324,"children":325},"ol",{},[326,337],{"type":48,"tag":182,"props":327,"children":328},{},[329,335],{"type":48,"tag":330,"props":331,"children":332},"strong",{},[333],{"type":54,"value":334},"Baseline first:",{"type":54,"value":336}," collect initial performance and system data, then choose an optimization workflow.",{"type":48,"tag":182,"props":338,"children":339},{},[340,345,347],{"type":48,"tag":330,"props":341,"children":342},{},[343],{"type":54,"value":344},"Optimize directly:",{"type":54,"value":346}," skip collection and choose one of:\n",{"type":48,"tag":178,"props":348,"children":349},{},[350,355,360],{"type":48,"tag":182,"props":351,"children":352},{},[353],{"type":54,"value":354},"remote transport selection for inter-node communication;",{"type":48,"tag":182,"props":356,"children":357},{},[358],{"type":54,"value":359},"NIC-to-PE mapping for topology-aware remote-transport placement;",{"type":48,"tag":182,"props":361,"children":362},{},[363],{"type":54,"value":364},"TMA assessment or enablement for eligible peer GPU\u002FNVLink kernel paths.",{"type":48,"tag":64,"props":366,"children":367},{},[368],{"type":54,"value":369},"Allow the user to collect data only and stop. Briefly distinguish the three optimization choices without recommending one from absent evidence.",{"type":48,"tag":199,"props":371,"children":373},{"id":372},"_3-sequence-explicit-multi-stage-workflows",[374],{"type":54,"value":375},"3. Sequence Explicit Multi-Stage Workflows",{"type":48,"tag":323,"props":377,"children":378},{},[379,384,389,394],{"type":48,"tag":182,"props":380,"children":381},{},[382],{"type":54,"value":383},"Invoke performance data collection first only when the user requests it, selects it from the menu, or explicitly chooses a baseline-first sequence. Let that skill obtain its required benchmark selection before any run.",{"type":48,"tag":182,"props":385,"children":386},{},[387],{"type":54,"value":388},"Invoke remote transport selection before NIC-to-PE mapping when both are requested. Pass the selected transport, version, provider, topology evidence, and unresolved conditions into the mapping workflow.",{"type":48,"tag":182,"props":390,"children":391},{},[392],{"type":54,"value":393},"Invoke TMA separately. Do not treat TMA as a remote-transport optimization or infer that inter-node benchmark evidence establishes TMA eligibility.",{"type":48,"tag":182,"props":395,"children":396},{},[397],{"type":54,"value":398},"Invoke no unrequested specialist. Let the active specialist resolve its own prerequisites instead of filling them in.",{"type":48,"tag":64,"props":400,"children":401},{},[402],{"type":54,"value":403},"After data collection completes, invoke an optimizer already selected by the user. If none was selected, present the three optimization choices again and wait.",{"type":48,"tag":199,"props":405,"children":407},{"id":406},"_4-preserve-the-handoff",[408],{"type":54,"value":409},"4. Preserve the Handoff",{"type":48,"tag":178,"props":411,"children":412},{},[413,418,423,428],{"type":48,"tag":182,"props":414,"children":415},{},[416],{"type":54,"value":417},"Pass the original request, explicit constraints, exact paths, commands, logs, configuration values, source scope, raw artifacts, resolved facts, and relevant prior specialist output.",{"type":48,"tag":182,"props":419,"children":420},{},[421],{"type":54,"value":422},"Preserve the user's transport settings and tuning constraints during collection and later comparison.",{"type":48,"tag":182,"props":424,"children":425},{},[426],{"type":54,"value":427},"Return the specialist's result without replacing its confidence, unknowns, validation status, or requested follow-up.",{"type":48,"tag":182,"props":429,"children":430},{},[431],{"type":54,"value":432},"Keep benchmark execution, system inspection, source edits, and configuration changes inside the selected specialist's authorization and guardrails.",{"type":48,"tag":199,"props":434,"children":436},{"id":435},"_5-handle-an-unavailable-specialist",[437],{"type":54,"value":438},"5. Handle an Unavailable Specialist",{"type":48,"tag":64,"props":440,"children":441},{},[442],{"type":54,"value":443},"If a required skill is unavailable, name it, explain which requested service cannot proceed, and ask the user to make it available or choose another service. Do not imitate or reconstruct the missing skill.",{"type":48,"tag":57,"props":445,"children":447},{"id":446},"limitations",[448],{"type":54,"value":449},"Limitations",{"type":48,"tag":178,"props":451,"children":452},{},[453,458,463,468,473,478],{"type":48,"tag":182,"props":454,"children":455},{},[456],{"type":54,"value":457},"Do not run benchmarks, collectors, launchers, or target applications directly from this router.",{"type":48,"tag":182,"props":459,"children":460},{},[461],{"type":54,"value":462},"Do not edit CUDA source, generate transport exports, or synthesize NIC mappings directly from this router.",{"type":48,"tag":182,"props":464,"children":465},{},[466],{"type":54,"value":467},"Do not make data collection an implicit prerequisite merely because an optimization might benefit from later measurement.",{"type":48,"tag":182,"props":469,"children":470},{},[471],{"type":54,"value":472},"Do not promise a performance improvement or declare a regression without the evidence required by the selected specialist.",{"type":48,"tag":182,"props":474,"children":475},{},[476],{"type":54,"value":477},"Do not treat TMA as useful for remote-network traffic; route it only as a separate peer GPU\u002FNVLink optimization workflow.",{"type":48,"tag":182,"props":479,"children":480},{},[481],{"type":54,"value":482},"Do not route unrelated NCCL, CUDA-only, operating-system, or general application-performance questions to these NVSHMEM skills.",{"type":48,"tag":57,"props":484,"children":486},{"id":485},"examples",[487],{"type":54,"value":488},"Examples",{"type":48,"tag":64,"props":490,"children":491},{},[492],{"type":54,"value":493},"Route a request to collect evidence before choosing an optimization:",{"type":48,"tag":495,"props":496,"children":501},"pre",{"className":497,"code":499,"language":54,"meta":500},[498],"language-text","User: Collect a baseline for our two-node NVSHMEM all-to-all before tuning it.\nAction: Invoke $nvshmem-collect-performance-data and preserve the two-node and all-to-all constraints.\n","",[502],{"type":48,"tag":114,"props":503,"children":504},{"__ignoreMap":500},[505],{"type":54,"value":499},{"type":48,"tag":64,"props":507,"children":508},{},[509],{"type":54,"value":510},"Route an explicit network-configuration request directly:",{"type":48,"tag":495,"props":512,"children":515},{"className":513,"code":514,"language":54,"meta":500},[498],"User: Which NVSHMEM remote transport should we use with this InfiniBand cluster?\nAction: Invoke $nvshmem-select-remote-transport because the request concerns an inter-node data path.\n",[516],{"type":48,"tag":114,"props":517,"children":518},{"__ignoreMap":500},[519],{"type":54,"value":514},{"type":48,"tag":57,"props":521,"children":523},{"id":522},"troubleshooting",[524],{"type":54,"value":525},"Troubleshooting",{"type":48,"tag":75,"props":527,"children":528},{},[529,550],{"type":48,"tag":79,"props":530,"children":531},{},[532],{"type":48,"tag":83,"props":533,"children":534},{},[535,540,545],{"type":48,"tag":87,"props":536,"children":537},{},[538],{"type":54,"value":539},"Error or condition",{"type":48,"tag":87,"props":541,"children":542},{},[543],{"type":54,"value":544},"Cause",{"type":48,"tag":87,"props":546,"children":547},{},[548],{"type":54,"value":549},"Solution",{"type":48,"tag":98,"props":551,"children":552},{},[553,571,589],{"type":48,"tag":83,"props":554,"children":555},{},[556,561,566],{"type":48,"tag":105,"props":557,"children":558},{},[559],{"type":54,"value":560},"No route can be selected",{"type":48,"tag":105,"props":562,"children":563},{},[564],{"type":54,"value":565},"The request says only to improve performance.",{"type":48,"tag":105,"props":567,"children":568},{},[569],{"type":54,"value":570},"Present the baseline-first and three direct-optimization options; wait for the user to choose.",{"type":48,"tag":83,"props":572,"children":573},{},[574,579,584],{"type":48,"tag":105,"props":575,"children":576},{},[577],{"type":54,"value":578},"A requested specialist is unavailable",{"type":48,"tag":105,"props":580,"children":581},{},[582],{"type":54,"value":583},"Its skill is not installed or enabled.",{"type":48,"tag":105,"props":585,"children":586},{},[587],{"type":54,"value":588},"Name the unavailable specialist and ask the user to enable it or choose another route.",{"type":48,"tag":83,"props":590,"children":591},{},[592,597,602],{"type":48,"tag":105,"props":593,"children":594},{},[595],{"type":54,"value":596},"Supplied settings conflict",{"type":48,"tag":105,"props":598,"children":599},{},[600],{"type":54,"value":601},"The request and earlier context specify different topology, transport, or constraints.",{"type":48,"tag":105,"props":603,"children":604},{},[605],{"type":54,"value":606},"Report the conflict and ask which value to preserve before handing off.",{"items":608,"total":693},[609,621,635,649,659,671,683],{"slug":610,"name":610,"fn":611,"description":612,"org":613,"tags":614,"stars":20,"repoUrl":21,"updatedAt":620},"nvshmem-collect-performance-data","collect NVSHMEM performance metrics","Collect and package NVSHMEM put\u002Fget bandwidth, latency, and other perftest results with system and topology evidence for performance sanity checks.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[615,618,619],{"name":616,"slug":617,"type":15},"Monitoring","monitoring",{"name":9,"slug":8,"type":15},{"name":13,"slug":14,"type":15},"2026-08-31T09:19:00.917887",{"slug":622,"name":622,"fn":623,"description":624,"org":625,"tags":626,"stars":20,"repoUrl":21,"updatedAt":634},"nvshmem-configure-nic-pe-mapping","configure NVSHMEM NIC-to-PE mappings","Recommend NVSHMEM NIC-to-PE mappings and environment exports. Use for HCA selection, multi-NIC configuration, or topology-based mapping diagnostics.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[627,630,633],{"name":628,"slug":629,"type":15},"Configuration","configuration",{"name":631,"slug":632,"type":15},"Networking","networking",{"name":9,"slug":8,"type":15},"2026-08-31T09:19:06.10294",{"slug":636,"name":636,"fn":637,"description":638,"org":639,"tags":640,"stars":20,"repoUrl":21,"updatedAt":648},"nvshmem-docs","fetch official NVSHMEM documentation","Find version-aware official NVSHMEM and NVSHMEM4Py documentation for releases, installation, APIs, runtime settings, transports, containers, and troubleshooting.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[641,644,645],{"name":642,"slug":643,"type":15},"Documentation","documentation",{"name":9,"slug":8,"type":15},{"name":646,"slug":647,"type":15},"Reference","reference","2026-08-31T09:19:00.214841",{"slug":650,"name":650,"fn":651,"description":652,"org":653,"tags":654,"stars":20,"repoUrl":21,"updatedAt":658},"nvshmem-enable-tma","optimize NVSHMEM CUDA kernels for TMA","Prepare or review NVSHMEM CUDA kernels for TMA SMEM registration and direct-SMEM transfers. Do not use for unrelated CUDA tuning.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[655,656,657],{"name":18,"slug":19,"type":15},{"name":9,"slug":8,"type":15},{"name":13,"slug":14,"type":15},"2026-08-31T09:18:50.449676",{"slug":660,"name":660,"fn":661,"description":662,"org":663,"tags":664,"stars":20,"repoUrl":21,"updatedAt":670},"nvshmem-get-started","onboard developers to NVSHMEM","Guide NVSHMEM beginners through fit assessment, mental models, first C\u002FC++ or Python NVSHMEM programs, compilation, launching, and next steps. Use for onboarding.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[665,666,667],{"name":642,"slug":643,"type":15},{"name":9,"slug":8,"type":15},{"name":668,"slug":669,"type":15},"Onboarding","onboarding","2026-08-31T09:18:55.48573",{"slug":672,"name":672,"fn":673,"description":674,"org":675,"tags":676,"stars":20,"repoUrl":21,"updatedAt":682},"nvshmem-install","install and validate NVSHMEM","Plan and validate NVSHMEM and NVSHMEM4Py installations. Use for package, container, or source deployments.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[677,680,681],{"name":678,"slug":679,"type":15},"Deployment","deployment",{"name":18,"slug":19,"type":15},{"name":9,"slug":8,"type":15},"2026-08-31T09:18:50.105521",{"slug":684,"name":684,"fn":685,"description":686,"org":687,"tags":688,"stars":20,"repoUrl":21,"updatedAt":692},"nvshmem-select-remote-transport","select NVSHMEM remote transport configurations","Select an NVSHMEM remote transport from target system and kernel evidence. Use for inter-node selection, compatibility checks, or configuration.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[689,690,691],{"name":628,"slug":629,"type":15},{"name":9,"slug":8,"type":15},{"name":13,"slug":14,"type":15},"2026-08-31T09:18:59.493722",9,{"items":695,"total":849},[696,712,727,738,750,764,777,791,804,815,829,838],{"slug":697,"name":697,"fn":698,"description":699,"org":700,"tags":701,"stars":709,"repoUrl":710,"updatedAt":711},"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},[702,703,706],{"name":642,"slug":643,"type":15},{"name":704,"slug":705,"type":15},"MCP","mcp",{"name":707,"slug":708,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-08-25T03:29:57.273192",{"slug":713,"name":713,"fn":714,"description":715,"org":716,"tags":717,"stars":724,"repoUrl":725,"updatedAt":726},"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},[718,721,722],{"name":719,"slug":720,"type":15},"Containers","containers",{"name":678,"slug":679,"type":15},{"name":723,"slug":30,"type":15},"Python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":728,"name":728,"fn":729,"description":730,"org":731,"tags":732,"stars":724,"repoUrl":725,"updatedAt":737},"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},[733,736],{"name":734,"slug":735,"type":15},"CI\u002FCD","ci-cd",{"name":678,"slug":679,"type":15},"2026-07-14T05:25:59.97109",{"slug":739,"name":739,"fn":740,"description":741,"org":742,"tags":743,"stars":724,"repoUrl":725,"updatedAt":749},"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},[744,745,746],{"name":734,"slug":735,"type":15},{"name":678,"slug":679,"type":15},{"name":747,"slug":748,"type":15},"GitHub","github","2026-08-28T14:38:16.959248",{"slug":751,"name":751,"fn":752,"description":753,"org":754,"tags":755,"stars":724,"repoUrl":725,"updatedAt":763},"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},[756,759,760],{"name":757,"slug":758,"type":15},"Debugging","debugging",{"name":747,"slug":748,"type":15},{"name":761,"slug":762,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":765,"name":765,"fn":766,"description":767,"org":768,"tags":769,"stars":724,"repoUrl":725,"updatedAt":776},"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},[770,773],{"name":771,"slug":772,"type":15},"Best Practices","best-practices",{"name":774,"slug":775,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":778,"name":778,"fn":779,"description":780,"org":781,"tags":782,"stars":724,"repoUrl":725,"updatedAt":790},"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, including the mechanical steps for transferring an existing pretrain_gpt.py launch script.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[783,786,789],{"name":784,"slug":785,"type":15},"Machine Learning","machine-learning",{"name":787,"slug":788,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-08-21T03:36:57.470256",{"slug":792,"name":792,"fn":793,"description":794,"org":795,"tags":796,"stars":724,"repoUrl":725,"updatedAt":803},"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},[797,800],{"name":798,"slug":799,"type":15},"QA","qa",{"name":801,"slug":802,"type":15},"Testing","testing","2026-07-14T05:25:53.673039",{"slug":805,"name":805,"fn":806,"description":807,"org":808,"tags":809,"stars":724,"repoUrl":725,"updatedAt":814},"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},[810,811],{"name":678,"slug":679,"type":15},{"name":812,"slug":813,"type":15},"Infrastructure","infrastructure","2026-07-14T05:25:49.362534",{"slug":816,"name":816,"fn":817,"description":818,"org":819,"tags":820,"stars":724,"repoUrl":725,"updatedAt":828},"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},[821,824,825],{"name":822,"slug":823,"type":15},"Code Review","code-review",{"name":747,"slug":748,"type":15},{"name":826,"slug":827,"type":15},"Pull Requests","pull-requests","2026-08-25T03:29:16.211287",{"slug":830,"name":830,"fn":831,"description":832,"org":833,"tags":834,"stars":724,"repoUrl":725,"updatedAt":837},"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},[835,836],{"name":798,"slug":799,"type":15},{"name":801,"slug":802,"type":15},"2026-07-14T05:25:54.928983",{"slug":839,"name":839,"fn":840,"description":841,"org":842,"tags":843,"stars":724,"repoUrl":725,"updatedAt":848},"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},[844,847],{"name":845,"slug":846,"type":15},"Automation","automation",{"name":734,"slug":735,"type":15},"2026-07-30T05:29:03.275638",563]