[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-nemo-mbridge-perf-hierarchical-context-parallel":3,"mdc-7402d6-key":31,"related-org-nvidia-nemo-mbridge-perf-hierarchical-context-parallel":1046,"related-repo-nvidia-nemo-mbridge-perf-hierarchical-context-parallel":1206},{"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":26,"sourceUrl":29,"mdContent":30},"nemo-mbridge-perf-hierarchical-context-parallel","enable hierarchical context parallelism in Megatron-Bridge","Operational guide for enabling hierarchical context parallelism in Megatron-Bridge, including config knobs, code anchors, pitfalls, and verification.",{"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",2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-07-14T05:26:53.953164","Apache-2.0",281,[],{"repoUrl":21,"stars":20,"forks":24,"topics":27,"description":28},[],"AI agent skills published by NVIDIA","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills\u002Ftree\u002FHEAD\u002Fskills\u002Fnemo-mbridge-perf-hierarchical-context-parallel","---\nname: nemo-mbridge-perf-hierarchical-context-parallel\ndescription: Operational guide for enabling hierarchical context parallelism in Megatron-Bridge, including config knobs, code anchors, pitfalls, and verification.\nlicense: Apache-2.0\nwhen_to_use: Scaling context parallelism beyond KV heads, or investigating a commit that changed CP config and caused OOM or a regression; 'hierarchical_context_parallel_sizes', 'a2a+p2p', 'hierarchical CP', 'CP beyond KV heads', 'multi-level CP'.\n---\n\n# Hierarchical Context Parallel Skill\n\nThis skill covers hierarchical context parallelism: nested context-parallel process\ngroups used by `cp_comm_type=\"a2a+p2p\"` and configured with\n`hierarchical_context_parallel_sizes`.\n\nFor what hierarchical CP is, when to use it, and the decision tree\n(`a2a+p2p` vs pure `a2a` vs `p2p`), see:\n\n- @docs\u002Ftraining\u002Fhierarchical-context-parallel.md\n- @skills\u002Fnemo-mbridge-perf-hierarchical-context-parallel\u002Fcard.yaml\n\n## Enablement\n\nMinimal Bridge override:\n\n```python\ncfg.model.context_parallel_size = 4\ncfg.model.cp_comm_type = \"a2a+p2p\"\ncfg.model.hierarchical_context_parallel_sizes = [2, 2]\ncfg.dist.use_decentralized_pg = False\n```\n\nRequired constraints:\n\n- `prod(hierarchical_context_parallel_sizes) == context_parallel_size`\n- `seq_length % (2 * context_parallel_size) == 0`\n- Transformer Engine `>= 1.12.0`\n\n## Code Anchors\n\nUpstream config and validation:\n\n```45:54:3rdparty\u002FMegatron-LM\u002Fmegatron\u002Fcore\u002Fmodel_parallel_config.py\ncontext_parallel_size: int = 1\n\"\"\"Splits network input along sequence dimension across GPU ranks.\"\"\"\n\nhierarchical_context_parallel_sizes: Optional[list[int]] = None\n\"\"\"Degrees of the hierarchical context parallelism. Users should provide a list to specify \n   the sizes for different levels. Taking the a2a+p2p cp comm type as example, it contains\n   groups of two levels, so the first value of the list indicates the group size of the a2a\n   communication type, and the second value indicates the group size of the p2p communication\n   type.\n\"\"\"\n```\n\n```428:433:3rdparty\u002FMegatron-LM\u002Fmegatron\u002Ftraining\u002Farguments.py\nif args.hierarchical_context_parallel_sizes:\n    from numpy import prod\n    assert args.context_parallel_size == prod(args.hierarchical_context_parallel_sizes)\nif \"a2a+p2p\" in args.cp_comm_type:\n    assert args.hierarchical_context_parallel_sizes is not None, \\\n    \"--hierarchical-context-parallel-sizes must be set when a2a+p2p is used in cp comm\"\n```\n\nBridge MPU path:\n\n```613:648:src\u002Fmegatron\u002Fbridge\u002Ftraining\u002Finitialize.py\nparallel_state.initialize_model_parallel(\n    ...\n    context_parallel_size=model_config.context_parallel_size,\n    hierarchical_context_parallel_sizes=model_config.hierarchical_context_parallel_sizes,\n    ...\n)\n...\nreturn ProcessGroupCollection.use_mpu_process_groups()\n```\n\nBridge decentralized-PG path:\n\n```503:524:src\u002Fmegatron\u002Fbridge\u002Ftraining\u002Finitialize.py\npg_collection = ProcessGroupCollection(\n    ...\n    cp=cp_pg,\n    tp_cp=tp_cp_pg,\n    hcp=None,\n    ep=ep_pg,\n    ...\n)\n```\n\n## Implementation Map\n\nThe code anchors above show the config declarations and argument validation.\n\n### Validation (MCore)\n\n`TransformerConfig.__post_init__` enforces that `a2a+p2p` requires HCP sizes and the product matches CP.\n\n### Process group creation\n\n`parallel_state.initialize_model_parallel` creates hierarchical CP sub-groups\nwhen HCP sizes are provided via `create_hierarchical_groups`. Bridge currently\ngets those groups through the MPU-backed `ProcessGroupCollection`.\n\n### TE integration\n\n`TEDotProductAttention` passes the hierarchical groups to Transformer Engine\nwhen `a2a+p2p` is used. Requires **Transformer Engine >= 1.12.0**.\n\n## Pitfalls\n\n1. **Bridge HCP is MPU-only today**: If `use_decentralized_pg=True`, Bridge initializes flat CP groups and leaves HCP unset.\n2. **No checked-in Bridge recipe** currently exercises HCP directly.\n3. **Single-GPU load helpers** clear `hierarchical_context_parallel_sizes`.\n4. **Silent broken training on old stacks**: If you use `a2a+p2p` without setting `hierarchical_context_parallel_sizes`, MCore now asserts. Older versions would silently disable CP communication, so each rank attended only to its local chunk and produced artificially high throughput with broken gradients.\n5. **Product must match**: `prod(hierarchical_context_parallel_sizes)` must exactly equal `context_parallel_size`. A mismatch triggers an assertion.\n6. **Verify in logs**: Look for the process group initialization output. You should see `HIERARCHICAL_CONTEXT_PARALLEL_GROUPS` being created. If you only see `CONTEXT_PARALLEL_GROUP`, HCP is not active.\n\n## Verification\n\nNo dedicated Bridge end-to-end test exists yet for HCP (see @skills\u002Fnemo-mbridge-perf-hierarchical-context-parallel\u002Fcard.yaml\n`follow_up_validation`). Use the existing unit tests and log inspection instead.\n\nRun the decentralized-PG unit test to confirm the flat-CP behavior is preserved:\n\n```bash\nuv run python -m pytest tests\u002Funit_tests\u002Ftraining\u002Ftest_decentralized_pg.py -q\n```\n\nFor a manual smoke check, launch a 4-GPU run with a small recipe and\n`cp_comm_type=a2a+p2p` plus `hierarchical_context_parallel_sizes=[2,2]`:\n\n```bash\nCUDA_VISIBLE_DEVICES=0,1,2,3 uv run python -m torch.distributed.run --nproc_per_node=4 \\\n  scripts\u002Ftraining\u002Frun_recipe.py \\\n  --recipe llama32_1b_pretrain_config \\\n  model.context_parallel_size=4 \\\n  model.cp_comm_type=a2a+p2p \\\n  \"model.hierarchical_context_parallel_sizes=[2,2]\" \\\n  train.train_iters=2\n```\n\nSuccess criteria:\n\n- Logs show `HIERARCHICAL_CONTEXT_PARALLEL_GROUPS` being created\n- Training completes at least one step without error\n- If you only see `CONTEXT_PARALLEL_GROUP`, HCP is not active\n",{"data":32,"body":34},{"name":4,"description":6,"license":23,"when_to_use":33},"Scaling context parallelism beyond KV heads, or investigating a commit that changed CP config and caused OOM or a regression; 'hierarchical_context_parallel_sizes', 'a2a+p2p', 'hierarchical CP', 'CP beyond KV heads', 'multi-level CP'.",{"type":35,"children":36},"root",[37,46,69,98,113,120,125,174,179,211,217,222,318,375,380,452,457,527,533,538,545,563,569,595,601,626,632,756,762,775,780,829,850,1003,1008,1040],{"type":38,"tag":39,"props":40,"children":42},"element","h1",{"id":41},"hierarchical-context-parallel-skill",[43],{"type":44,"value":45},"text","Hierarchical Context Parallel Skill",{"type":38,"tag":47,"props":48,"children":49},"p",{},[50,52,59,61,67],{"type":44,"value":51},"This skill covers hierarchical context parallelism: nested context-parallel process\ngroups used by ",{"type":38,"tag":53,"props":54,"children":56},"code",{"className":55},[],[57],{"type":44,"value":58},"cp_comm_type=\"a2a+p2p\"",{"type":44,"value":60}," and configured with\n",{"type":38,"tag":53,"props":62,"children":64},{"className":63},[],[65],{"type":44,"value":66},"hierarchical_context_parallel_sizes",{"type":44,"value":68},".",{"type":38,"tag":47,"props":70,"children":71},{},[72,74,80,82,88,90,96],{"type":44,"value":73},"For what hierarchical CP is, when to use it, and the decision tree\n(",{"type":38,"tag":53,"props":75,"children":77},{"className":76},[],[78],{"type":44,"value":79},"a2a+p2p",{"type":44,"value":81}," vs pure ",{"type":38,"tag":53,"props":83,"children":85},{"className":84},[],[86],{"type":44,"value":87},"a2a",{"type":44,"value":89}," vs ",{"type":38,"tag":53,"props":91,"children":93},{"className":92},[],[94],{"type":44,"value":95},"p2p",{"type":44,"value":97},"), see:",{"type":38,"tag":99,"props":100,"children":101},"ul",{},[102,108],{"type":38,"tag":103,"props":104,"children":105},"li",{},[106],{"type":44,"value":107},"@docs\u002Ftraining\u002Fhierarchical-context-parallel.md",{"type":38,"tag":103,"props":109,"children":110},{},[111],{"type":44,"value":112},"@skills\u002Fnemo-mbridge-perf-hierarchical-context-parallel\u002Fcard.yaml",{"type":38,"tag":114,"props":115,"children":117},"h2",{"id":116},"enablement",[118],{"type":44,"value":119},"Enablement",{"type":38,"tag":47,"props":121,"children":122},{},[123],{"type":44,"value":124},"Minimal Bridge override:",{"type":38,"tag":126,"props":127,"children":132},"pre",{"className":128,"code":129,"language":130,"meta":131,"style":131},"language-python shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","cfg.model.context_parallel_size = 4\ncfg.model.cp_comm_type = \"a2a+p2p\"\ncfg.model.hierarchical_context_parallel_sizes = [2, 2]\ncfg.dist.use_decentralized_pg = False\n","python","",[133],{"type":38,"tag":53,"props":134,"children":135},{"__ignoreMap":131},[136,147,156,165],{"type":38,"tag":137,"props":138,"children":141},"span",{"class":139,"line":140},"line",1,[142],{"type":38,"tag":137,"props":143,"children":144},{},[145],{"type":44,"value":146},"cfg.model.context_parallel_size = 4\n",{"type":38,"tag":137,"props":148,"children":150},{"class":139,"line":149},2,[151],{"type":38,"tag":137,"props":152,"children":153},{},[154],{"type":44,"value":155},"cfg.model.cp_comm_type = \"a2a+p2p\"\n",{"type":38,"tag":137,"props":157,"children":159},{"class":139,"line":158},3,[160],{"type":38,"tag":137,"props":161,"children":162},{},[163],{"type":44,"value":164},"cfg.model.hierarchical_context_parallel_sizes = [2, 2]\n",{"type":38,"tag":137,"props":166,"children":168},{"class":139,"line":167},4,[169],{"type":38,"tag":137,"props":170,"children":171},{},[172],{"type":44,"value":173},"cfg.dist.use_decentralized_pg = False\n",{"type":38,"tag":47,"props":175,"children":176},{},[177],{"type":44,"value":178},"Required constraints:",{"type":38,"tag":99,"props":180,"children":181},{},[182,191,200],{"type":38,"tag":103,"props":183,"children":184},{},[185],{"type":38,"tag":53,"props":186,"children":188},{"className":187},[],[189],{"type":44,"value":190},"prod(hierarchical_context_parallel_sizes) == context_parallel_size",{"type":38,"tag":103,"props":192,"children":193},{},[194],{"type":38,"tag":53,"props":195,"children":197},{"className":196},[],[198],{"type":44,"value":199},"seq_length % (2 * context_parallel_size) == 0",{"type":38,"tag":103,"props":201,"children":202},{},[203,205],{"type":44,"value":204},"Transformer Engine ",{"type":38,"tag":53,"props":206,"children":208},{"className":207},[],[209],{"type":44,"value":210},">= 1.12.0",{"type":38,"tag":114,"props":212,"children":214},{"id":213},"code-anchors",[215],{"type":44,"value":216},"Code Anchors",{"type":38,"tag":47,"props":218,"children":219},{},[220],{"type":44,"value":221},"Upstream config and validation:",{"type":38,"tag":126,"props":223,"children":227},{"className":224,"code":225,"language":226,"meta":131,"style":131},"language-45:54:3rdparty\u002FMegatron-LM\u002Fmegatron\u002Fcore\u002Fmodel_parallel_config.py shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","context_parallel_size: int = 1\n\"\"\"Splits network input along sequence dimension across GPU ranks.\"\"\"\n\nhierarchical_context_parallel_sizes: Optional[list[int]] = None\n\"\"\"Degrees of the hierarchical context parallelism. Users should provide a list to specify \n   the sizes for different levels. Taking the a2a+p2p cp comm type as example, it contains\n   groups of two levels, so the first value of the list indicates the group size of the a2a\n   communication type, and the second value indicates the group size of the p2p communication\n   type.\n\"\"\"\n","45:54:3rdparty\u002FMegatron-LM\u002Fmegatron\u002Fcore\u002Fmodel_parallel_config.py",[228],{"type":38,"tag":53,"props":229,"children":230},{"__ignoreMap":131},[231,239,247,256,264,273,282,291,300,309],{"type":38,"tag":137,"props":232,"children":233},{"class":139,"line":140},[234],{"type":38,"tag":137,"props":235,"children":236},{},[237],{"type":44,"value":238},"context_parallel_size: int = 1\n",{"type":38,"tag":137,"props":240,"children":241},{"class":139,"line":149},[242],{"type":38,"tag":137,"props":243,"children":244},{},[245],{"type":44,"value":246},"\"\"\"Splits network input along sequence dimension across GPU ranks.\"\"\"\n",{"type":38,"tag":137,"props":248,"children":249},{"class":139,"line":158},[250],{"type":38,"tag":137,"props":251,"children":253},{"emptyLinePlaceholder":252},true,[254],{"type":44,"value":255},"\n",{"type":38,"tag":137,"props":257,"children":258},{"class":139,"line":167},[259],{"type":38,"tag":137,"props":260,"children":261},{},[262],{"type":44,"value":263},"hierarchical_context_parallel_sizes: Optional[list[int]] = None\n",{"type":38,"tag":137,"props":265,"children":267},{"class":139,"line":266},5,[268],{"type":38,"tag":137,"props":269,"children":270},{},[271],{"type":44,"value":272},"\"\"\"Degrees of the hierarchical context parallelism. Users should provide a list to specify \n",{"type":38,"tag":137,"props":274,"children":276},{"class":139,"line":275},6,[277],{"type":38,"tag":137,"props":278,"children":279},{},[280],{"type":44,"value":281},"   the sizes for different levels. Taking the a2a+p2p cp comm type as example, it contains\n",{"type":38,"tag":137,"props":283,"children":285},{"class":139,"line":284},7,[286],{"type":38,"tag":137,"props":287,"children":288},{},[289],{"type":44,"value":290},"   groups of two levels, so the first value of the list indicates the group size of the a2a\n",{"type":38,"tag":137,"props":292,"children":294},{"class":139,"line":293},8,[295],{"type":38,"tag":137,"props":296,"children":297},{},[298],{"type":44,"value":299},"   communication type, and the second value indicates the group size of the p2p communication\n",{"type":38,"tag":137,"props":301,"children":303},{"class":139,"line":302},9,[304],{"type":38,"tag":137,"props":305,"children":306},{},[307],{"type":44,"value":308},"   type.\n",{"type":38,"tag":137,"props":310,"children":312},{"class":139,"line":311},10,[313],{"type":38,"tag":137,"props":314,"children":315},{},[316],{"type":44,"value":317},"\"\"\"\n",{"type":38,"tag":126,"props":319,"children":323},{"className":320,"code":321,"language":322,"meta":131,"style":131},"language-428:433:3rdparty\u002FMegatron-LM\u002Fmegatron\u002Ftraining\u002Farguments.py shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","if args.hierarchical_context_parallel_sizes:\n    from numpy import prod\n    assert args.context_parallel_size == prod(args.hierarchical_context_parallel_sizes)\nif \"a2a+p2p\" in args.cp_comm_type:\n    assert args.hierarchical_context_parallel_sizes is not None, \\\n    \"--hierarchical-context-parallel-sizes must be set when a2a+p2p is used in cp comm\"\n","428:433:3rdparty\u002FMegatron-LM\u002Fmegatron\u002Ftraining\u002Farguments.py",[324],{"type":38,"tag":53,"props":325,"children":326},{"__ignoreMap":131},[327,335,343,351,359,367],{"type":38,"tag":137,"props":328,"children":329},{"class":139,"line":140},[330],{"type":38,"tag":137,"props":331,"children":332},{},[333],{"type":44,"value":334},"if args.hierarchical_context_parallel_sizes:\n",{"type":38,"tag":137,"props":336,"children":337},{"class":139,"line":149},[338],{"type":38,"tag":137,"props":339,"children":340},{},[341],{"type":44,"value":342},"    from numpy import prod\n",{"type":38,"tag":137,"props":344,"children":345},{"class":139,"line":158},[346],{"type":38,"tag":137,"props":347,"children":348},{},[349],{"type":44,"value":350},"    assert args.context_parallel_size == prod(args.hierarchical_context_parallel_sizes)\n",{"type":38,"tag":137,"props":352,"children":353},{"class":139,"line":167},[354],{"type":38,"tag":137,"props":355,"children":356},{},[357],{"type":44,"value":358},"if \"a2a+p2p\" in args.cp_comm_type:\n",{"type":38,"tag":137,"props":360,"children":361},{"class":139,"line":266},[362],{"type":38,"tag":137,"props":363,"children":364},{},[365],{"type":44,"value":366},"    assert args.hierarchical_context_parallel_sizes is not None, \\\n",{"type":38,"tag":137,"props":368,"children":369},{"class":139,"line":275},[370],{"type":38,"tag":137,"props":371,"children":372},{},[373],{"type":44,"value":374},"    \"--hierarchical-context-parallel-sizes must be set when a2a+p2p is used in cp comm\"\n",{"type":38,"tag":47,"props":376,"children":377},{},[378],{"type":44,"value":379},"Bridge MPU path:",{"type":38,"tag":126,"props":381,"children":385},{"className":382,"code":383,"language":384,"meta":131,"style":131},"language-613:648:src\u002Fmegatron\u002Fbridge\u002Ftraining\u002Finitialize.py shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","parallel_state.initialize_model_parallel(\n    ...\n    context_parallel_size=model_config.context_parallel_size,\n    hierarchical_context_parallel_sizes=model_config.hierarchical_context_parallel_sizes,\n    ...\n)\n...\nreturn ProcessGroupCollection.use_mpu_process_groups()\n","613:648:src\u002Fmegatron\u002Fbridge\u002Ftraining\u002Finitialize.py",[386],{"type":38,"tag":53,"props":387,"children":388},{"__ignoreMap":131},[389,397,405,413,421,428,436,444],{"type":38,"tag":137,"props":390,"children":391},{"class":139,"line":140},[392],{"type":38,"tag":137,"props":393,"children":394},{},[395],{"type":44,"value":396},"parallel_state.initialize_model_parallel(\n",{"type":38,"tag":137,"props":398,"children":399},{"class":139,"line":149},[400],{"type":38,"tag":137,"props":401,"children":402},{},[403],{"type":44,"value":404},"    ...\n",{"type":38,"tag":137,"props":406,"children":407},{"class":139,"line":158},[408],{"type":38,"tag":137,"props":409,"children":410},{},[411],{"type":44,"value":412},"    context_parallel_size=model_config.context_parallel_size,\n",{"type":38,"tag":137,"props":414,"children":415},{"class":139,"line":167},[416],{"type":38,"tag":137,"props":417,"children":418},{},[419],{"type":44,"value":420},"    hierarchical_context_parallel_sizes=model_config.hierarchical_context_parallel_sizes,\n",{"type":38,"tag":137,"props":422,"children":423},{"class":139,"line":266},[424],{"type":38,"tag":137,"props":425,"children":426},{},[427],{"type":44,"value":404},{"type":38,"tag":137,"props":429,"children":430},{"class":139,"line":275},[431],{"type":38,"tag":137,"props":432,"children":433},{},[434],{"type":44,"value":435},")\n",{"type":38,"tag":137,"props":437,"children":438},{"class":139,"line":284},[439],{"type":38,"tag":137,"props":440,"children":441},{},[442],{"type":44,"value":443},"...\n",{"type":38,"tag":137,"props":445,"children":446},{"class":139,"line":293},[447],{"type":38,"tag":137,"props":448,"children":449},{},[450],{"type":44,"value":451},"return ProcessGroupCollection.use_mpu_process_groups()\n",{"type":38,"tag":47,"props":453,"children":454},{},[455],{"type":44,"value":456},"Bridge decentralized-PG path:",{"type":38,"tag":126,"props":458,"children":462},{"className":459,"code":460,"language":461,"meta":131,"style":131},"language-503:524:src\u002Fmegatron\u002Fbridge\u002Ftraining\u002Finitialize.py shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","pg_collection = ProcessGroupCollection(\n    ...\n    cp=cp_pg,\n    tp_cp=tp_cp_pg,\n    hcp=None,\n    ep=ep_pg,\n    ...\n)\n","503:524:src\u002Fmegatron\u002Fbridge\u002Ftraining\u002Finitialize.py",[463],{"type":38,"tag":53,"props":464,"children":465},{"__ignoreMap":131},[466,474,481,489,497,505,513,520],{"type":38,"tag":137,"props":467,"children":468},{"class":139,"line":140},[469],{"type":38,"tag":137,"props":470,"children":471},{},[472],{"type":44,"value":473},"pg_collection = ProcessGroupCollection(\n",{"type":38,"tag":137,"props":475,"children":476},{"class":139,"line":149},[477],{"type":38,"tag":137,"props":478,"children":479},{},[480],{"type":44,"value":404},{"type":38,"tag":137,"props":482,"children":483},{"class":139,"line":158},[484],{"type":38,"tag":137,"props":485,"children":486},{},[487],{"type":44,"value":488},"    cp=cp_pg,\n",{"type":38,"tag":137,"props":490,"children":491},{"class":139,"line":167},[492],{"type":38,"tag":137,"props":493,"children":494},{},[495],{"type":44,"value":496},"    tp_cp=tp_cp_pg,\n",{"type":38,"tag":137,"props":498,"children":499},{"class":139,"line":266},[500],{"type":38,"tag":137,"props":501,"children":502},{},[503],{"type":44,"value":504},"    hcp=None,\n",{"type":38,"tag":137,"props":506,"children":507},{"class":139,"line":275},[508],{"type":38,"tag":137,"props":509,"children":510},{},[511],{"type":44,"value":512},"    ep=ep_pg,\n",{"type":38,"tag":137,"props":514,"children":515},{"class":139,"line":284},[516],{"type":38,"tag":137,"props":517,"children":518},{},[519],{"type":44,"value":404},{"type":38,"tag":137,"props":521,"children":522},{"class":139,"line":293},[523],{"type":38,"tag":137,"props":524,"children":525},{},[526],{"type":44,"value":435},{"type":38,"tag":114,"props":528,"children":530},{"id":529},"implementation-map",[531],{"type":44,"value":532},"Implementation Map",{"type":38,"tag":47,"props":534,"children":535},{},[536],{"type":44,"value":537},"The code anchors above show the config declarations and argument validation.",{"type":38,"tag":539,"props":540,"children":542},"h3",{"id":541},"validation-mcore",[543],{"type":44,"value":544},"Validation (MCore)",{"type":38,"tag":47,"props":546,"children":547},{},[548,554,556,561],{"type":38,"tag":53,"props":549,"children":551},{"className":550},[],[552],{"type":44,"value":553},"TransformerConfig.__post_init__",{"type":44,"value":555}," enforces that ",{"type":38,"tag":53,"props":557,"children":559},{"className":558},[],[560],{"type":44,"value":79},{"type":44,"value":562}," requires HCP sizes and the product matches CP.",{"type":38,"tag":539,"props":564,"children":566},{"id":565},"process-group-creation",[567],{"type":44,"value":568},"Process group creation",{"type":38,"tag":47,"props":570,"children":571},{},[572,578,580,586,588,594],{"type":38,"tag":53,"props":573,"children":575},{"className":574},[],[576],{"type":44,"value":577},"parallel_state.initialize_model_parallel",{"type":44,"value":579}," creates hierarchical CP sub-groups\nwhen HCP sizes are provided via ",{"type":38,"tag":53,"props":581,"children":583},{"className":582},[],[584],{"type":44,"value":585},"create_hierarchical_groups",{"type":44,"value":587},". Bridge currently\ngets those groups through the MPU-backed ",{"type":38,"tag":53,"props":589,"children":591},{"className":590},[],[592],{"type":44,"value":593},"ProcessGroupCollection",{"type":44,"value":68},{"type":38,"tag":539,"props":596,"children":598},{"id":597},"te-integration",[599],{"type":44,"value":600},"TE integration",{"type":38,"tag":47,"props":602,"children":603},{},[604,610,612,617,619,625],{"type":38,"tag":53,"props":605,"children":607},{"className":606},[],[608],{"type":44,"value":609},"TEDotProductAttention",{"type":44,"value":611}," passes the hierarchical groups to Transformer Engine\nwhen ",{"type":38,"tag":53,"props":613,"children":615},{"className":614},[],[616],{"type":44,"value":79},{"type":44,"value":618}," is used. Requires ",{"type":38,"tag":620,"props":621,"children":622},"strong",{},[623],{"type":44,"value":624},"Transformer Engine >= 1.12.0",{"type":44,"value":68},{"type":38,"tag":114,"props":627,"children":629},{"id":628},"pitfalls",[630],{"type":44,"value":631},"Pitfalls",{"type":38,"tag":633,"props":634,"children":635},"ol",{},[636,654,664,680,704,730],{"type":38,"tag":103,"props":637,"children":638},{},[639,644,646,652],{"type":38,"tag":620,"props":640,"children":641},{},[642],{"type":44,"value":643},"Bridge HCP is MPU-only today",{"type":44,"value":645},": If ",{"type":38,"tag":53,"props":647,"children":649},{"className":648},[],[650],{"type":44,"value":651},"use_decentralized_pg=True",{"type":44,"value":653},", Bridge initializes flat CP groups and leaves HCP unset.",{"type":38,"tag":103,"props":655,"children":656},{},[657,662],{"type":38,"tag":620,"props":658,"children":659},{},[660],{"type":44,"value":661},"No checked-in Bridge recipe",{"type":44,"value":663}," currently exercises HCP directly.",{"type":38,"tag":103,"props":665,"children":666},{},[667,672,674,679],{"type":38,"tag":620,"props":668,"children":669},{},[670],{"type":44,"value":671},"Single-GPU load helpers",{"type":44,"value":673}," clear ",{"type":38,"tag":53,"props":675,"children":677},{"className":676},[],[678],{"type":44,"value":66},{"type":44,"value":68},{"type":38,"tag":103,"props":681,"children":682},{},[683,688,690,695,697,702],{"type":38,"tag":620,"props":684,"children":685},{},[686],{"type":44,"value":687},"Silent broken training on old stacks",{"type":44,"value":689},": If you use ",{"type":38,"tag":53,"props":691,"children":693},{"className":692},[],[694],{"type":44,"value":79},{"type":44,"value":696}," without setting ",{"type":38,"tag":53,"props":698,"children":700},{"className":699},[],[701],{"type":44,"value":66},{"type":44,"value":703},", MCore now asserts. Older versions would silently disable CP communication, so each rank attended only to its local chunk and produced artificially high throughput with broken gradients.",{"type":38,"tag":103,"props":705,"children":706},{},[707,712,714,720,722,728],{"type":38,"tag":620,"props":708,"children":709},{},[710],{"type":44,"value":711},"Product must match",{"type":44,"value":713},": ",{"type":38,"tag":53,"props":715,"children":717},{"className":716},[],[718],{"type":44,"value":719},"prod(hierarchical_context_parallel_sizes)",{"type":44,"value":721}," must exactly equal ",{"type":38,"tag":53,"props":723,"children":725},{"className":724},[],[726],{"type":44,"value":727},"context_parallel_size",{"type":44,"value":729},". A mismatch triggers an assertion.",{"type":38,"tag":103,"props":731,"children":732},{},[733,738,740,746,748,754],{"type":38,"tag":620,"props":734,"children":735},{},[736],{"type":44,"value":737},"Verify in logs",{"type":44,"value":739},": Look for the process group initialization output. You should see ",{"type":38,"tag":53,"props":741,"children":743},{"className":742},[],[744],{"type":44,"value":745},"HIERARCHICAL_CONTEXT_PARALLEL_GROUPS",{"type":44,"value":747}," being created. If you only see ",{"type":38,"tag":53,"props":749,"children":751},{"className":750},[],[752],{"type":44,"value":753},"CONTEXT_PARALLEL_GROUP",{"type":44,"value":755},", HCP is not active.",{"type":38,"tag":114,"props":757,"children":759},{"id":758},"verification",[760],{"type":44,"value":761},"Verification",{"type":38,"tag":47,"props":763,"children":764},{},[765,767,773],{"type":44,"value":766},"No dedicated Bridge end-to-end test exists yet for HCP (see @skills\u002Fnemo-mbridge-perf-hierarchical-context-parallel\u002Fcard.yaml\n",{"type":38,"tag":53,"props":768,"children":770},{"className":769},[],[771],{"type":44,"value":772},"follow_up_validation",{"type":44,"value":774},"). Use the existing unit tests and log inspection instead.",{"type":38,"tag":47,"props":776,"children":777},{},[778],{"type":44,"value":779},"Run the decentralized-PG unit test to confirm the flat-CP behavior is preserved:",{"type":38,"tag":126,"props":781,"children":785},{"className":782,"code":783,"language":784,"meta":131,"style":131},"language-bash shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","uv run python -m pytest tests\u002Funit_tests\u002Ftraining\u002Ftest_decentralized_pg.py -q\n","bash",[786],{"type":38,"tag":53,"props":787,"children":788},{"__ignoreMap":131},[789],{"type":38,"tag":137,"props":790,"children":791},{"class":139,"line":140},[792,798,804,809,814,819,824],{"type":38,"tag":137,"props":793,"children":795},{"style":794},"--shiki-light:#E2931D;--shiki-default:#FFCB6B;--shiki-dark:#FFCB6B",[796],{"type":44,"value":797},"uv",{"type":38,"tag":137,"props":799,"children":801},{"style":800},"--shiki-light:#91B859;--shiki-default:#C3E88D;--shiki-dark:#C3E88D",[802],{"type":44,"value":803}," run",{"type":38,"tag":137,"props":805,"children":806},{"style":800},[807],{"type":44,"value":808}," python",{"type":38,"tag":137,"props":810,"children":811},{"style":800},[812],{"type":44,"value":813}," -m",{"type":38,"tag":137,"props":815,"children":816},{"style":800},[817],{"type":44,"value":818}," pytest",{"type":38,"tag":137,"props":820,"children":821},{"style":800},[822],{"type":44,"value":823}," tests\u002Funit_tests\u002Ftraining\u002Ftest_decentralized_pg.py",{"type":38,"tag":137,"props":825,"children":826},{"style":800},[827],{"type":44,"value":828}," -q\n",{"type":38,"tag":47,"props":830,"children":831},{},[832,834,840,842,848],{"type":44,"value":833},"For a manual smoke check, launch a 4-GPU run with a small recipe and\n",{"type":38,"tag":53,"props":835,"children":837},{"className":836},[],[838],{"type":44,"value":839},"cp_comm_type=a2a+p2p",{"type":44,"value":841}," plus ",{"type":38,"tag":53,"props":843,"children":845},{"className":844},[],[846],{"type":44,"value":847},"hierarchical_context_parallel_sizes=[2,2]",{"type":44,"value":849},":",{"type":38,"tag":126,"props":851,"children":853},{"className":782,"code":852,"language":784,"meta":131,"style":131},"CUDA_VISIBLE_DEVICES=0,1,2,3 uv run python -m torch.distributed.run --nproc_per_node=4 \\\n  scripts\u002Ftraining\u002Frun_recipe.py \\\n  --recipe llama32_1b_pretrain_config \\\n  model.context_parallel_size=4 \\\n  model.cp_comm_type=a2a+p2p \\\n  \"model.hierarchical_context_parallel_sizes=[2,2]\" \\\n  train.train_iters=2\n",[854],{"type":38,"tag":53,"props":855,"children":856},{"__ignoreMap":131},[857,909,921,938,956,968,990],{"type":38,"tag":137,"props":858,"children":859},{"class":139,"line":140},[860,866,872,877,882,886,890,894,899,904],{"type":38,"tag":137,"props":861,"children":863},{"style":862},"--shiki-light:#90A4AE;--shiki-default:#EEFFFF;--shiki-dark:#BABED8",[864],{"type":44,"value":865},"CUDA_VISIBLE_DEVICES",{"type":38,"tag":137,"props":867,"children":869},{"style":868},"--shiki-light:#39ADB5;--shiki-default:#89DDFF;--shiki-dark:#89DDFF",[870],{"type":44,"value":871},"=",{"type":38,"tag":137,"props":873,"children":874},{"style":800},[875],{"type":44,"value":876},"0,1,2,3",{"type":38,"tag":137,"props":878,"children":879},{"style":794},[880],{"type":44,"value":881}," uv",{"type":38,"tag":137,"props":883,"children":884},{"style":800},[885],{"type":44,"value":803},{"type":38,"tag":137,"props":887,"children":888},{"style":800},[889],{"type":44,"value":808},{"type":38,"tag":137,"props":891,"children":892},{"style":800},[893],{"type":44,"value":813},{"type":38,"tag":137,"props":895,"children":896},{"style":800},[897],{"type":44,"value":898}," torch.distributed.run",{"type":38,"tag":137,"props":900,"children":901},{"style":800},[902],{"type":44,"value":903}," --nproc_per_node=4",{"type":38,"tag":137,"props":905,"children":906},{"style":862},[907],{"type":44,"value":908}," \\\n",{"type":38,"tag":137,"props":910,"children":911},{"class":139,"line":149},[912,917],{"type":38,"tag":137,"props":913,"children":914},{"style":800},[915],{"type":44,"value":916},"  scripts\u002Ftraining\u002Frun_recipe.py",{"type":38,"tag":137,"props":918,"children":919},{"style":862},[920],{"type":44,"value":908},{"type":38,"tag":137,"props":922,"children":923},{"class":139,"line":158},[924,929,934],{"type":38,"tag":137,"props":925,"children":926},{"style":800},[927],{"type":44,"value":928},"  --recipe",{"type":38,"tag":137,"props":930,"children":931},{"style":800},[932],{"type":44,"value":933}," llama32_1b_pretrain_config",{"type":38,"tag":137,"props":935,"children":936},{"style":862},[937],{"type":44,"value":908},{"type":38,"tag":137,"props":939,"children":940},{"class":139,"line":167},[941,946,952],{"type":38,"tag":137,"props":942,"children":943},{"style":800},[944],{"type":44,"value":945},"  model.context_parallel_size=",{"type":38,"tag":137,"props":947,"children":949},{"style":948},"--shiki-light:#F76D47;--shiki-default:#F78C6C;--shiki-dark:#F78C6C",[950],{"type":44,"value":951},"4",{"type":38,"tag":137,"props":953,"children":954},{"style":862},[955],{"type":44,"value":908},{"type":38,"tag":137,"props":957,"children":958},{"class":139,"line":266},[959,964],{"type":38,"tag":137,"props":960,"children":961},{"style":800},[962],{"type":44,"value":963},"  model.cp_comm_type=a2a+p2p",{"type":38,"tag":137,"props":965,"children":966},{"style":862},[967],{"type":44,"value":908},{"type":38,"tag":137,"props":969,"children":970},{"class":139,"line":275},[971,976,981,986],{"type":38,"tag":137,"props":972,"children":973},{"style":868},[974],{"type":44,"value":975},"  \"",{"type":38,"tag":137,"props":977,"children":978},{"style":800},[979],{"type":44,"value":980},"model.hierarchical_context_parallel_sizes=[2,2]",{"type":38,"tag":137,"props":982,"children":983},{"style":868},[984],{"type":44,"value":985},"\"",{"type":38,"tag":137,"props":987,"children":988},{"style":862},[989],{"type":44,"value":908},{"type":38,"tag":137,"props":991,"children":992},{"class":139,"line":284},[993,998],{"type":38,"tag":137,"props":994,"children":995},{"style":800},[996],{"type":44,"value":997},"  train.train_iters=",{"type":38,"tag":137,"props":999,"children":1000},{"style":948},[1001],{"type":44,"value":1002},"2\n",{"type":38,"tag":47,"props":1004,"children":1005},{},[1006],{"type":44,"value":1007},"Success criteria:",{"type":38,"tag":99,"props":1009,"children":1010},{},[1011,1023,1028],{"type":38,"tag":103,"props":1012,"children":1013},{},[1014,1016,1021],{"type":44,"value":1015},"Logs show ",{"type":38,"tag":53,"props":1017,"children":1019},{"className":1018},[],[1020],{"type":44,"value":745},{"type":44,"value":1022}," being created",{"type":38,"tag":103,"props":1024,"children":1025},{},[1026],{"type":44,"value":1027},"Training completes at least one step without error",{"type":38,"tag":103,"props":1029,"children":1030},{},[1031,1033,1038],{"type":44,"value":1032},"If you only see ",{"type":38,"tag":53,"props":1034,"children":1036},{"className":1035},[],[1037],{"type":44,"value":753},{"type":44,"value":1039},", HCP is not active",{"type":38,"tag":1041,"props":1042,"children":1043},"style",{},[1044],{"type":44,"value":1045},"html .light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html.light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"items":1047,"total":1205},[1048,1066,1083,1094,1106,1120,1133,1147,1160,1171,1185,1194],{"slug":1049,"name":1049,"fn":1050,"description":1051,"org":1052,"tags":1053,"stars":1063,"repoUrl":1064,"updatedAt":1065},"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},[1054,1057,1060],{"name":1055,"slug":1056,"type":15},"Documentation","documentation",{"name":1058,"slug":1059,"type":15},"MCP","mcp",{"name":1061,"slug":1062,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":1067,"name":1067,"fn":1068,"description":1069,"org":1070,"tags":1071,"stars":1080,"repoUrl":1081,"updatedAt":1082},"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},[1072,1075,1078],{"name":1073,"slug":1074,"type":15},"Containers","containers",{"name":1076,"slug":1077,"type":15},"Deployment","deployment",{"name":1079,"slug":130,"type":15},"Python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":1084,"name":1084,"fn":1085,"description":1086,"org":1087,"tags":1088,"stars":1080,"repoUrl":1081,"updatedAt":1093},"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},[1089,1092],{"name":1090,"slug":1091,"type":15},"CI\u002FCD","ci-cd",{"name":1076,"slug":1077,"type":15},"2026-07-14T05:25:59.97109",{"slug":1095,"name":1095,"fn":1096,"description":1097,"org":1098,"tags":1099,"stars":1080,"repoUrl":1081,"updatedAt":1105},"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},[1100,1101,1102],{"name":1090,"slug":1091,"type":15},{"name":1076,"slug":1077,"type":15},{"name":1103,"slug":1104,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":1107,"name":1107,"fn":1108,"description":1109,"org":1110,"tags":1111,"stars":1080,"repoUrl":1081,"updatedAt":1119},"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},[1112,1115,1116],{"name":1113,"slug":1114,"type":15},"Debugging","debugging",{"name":1103,"slug":1104,"type":15},{"name":1117,"slug":1118,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":1121,"name":1121,"fn":1122,"description":1123,"org":1124,"tags":1125,"stars":1080,"repoUrl":1081,"updatedAt":1132},"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},[1126,1129],{"name":1127,"slug":1128,"type":15},"Best Practices","best-practices",{"name":1130,"slug":1131,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":1134,"name":1134,"fn":1135,"description":1136,"org":1137,"tags":1138,"stars":1080,"repoUrl":1081,"updatedAt":1146},"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},[1139,1142,1145],{"name":1140,"slug":1141,"type":15},"Machine Learning","machine-learning",{"name":1143,"slug":1144,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":1148,"name":1148,"fn":1149,"description":1150,"org":1151,"tags":1152,"stars":1080,"repoUrl":1081,"updatedAt":1159},"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},[1153,1156],{"name":1154,"slug":1155,"type":15},"QA","qa",{"name":1157,"slug":1158,"type":15},"Testing","testing","2026-07-14T05:25:53.673039",{"slug":1161,"name":1161,"fn":1162,"description":1163,"org":1164,"tags":1165,"stars":1080,"repoUrl":1081,"updatedAt":1170},"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},[1166,1167],{"name":1076,"slug":1077,"type":15},{"name":1168,"slug":1169,"type":15},"Infrastructure","infrastructure","2026-07-14T05:25:49.362534",{"slug":1172,"name":1172,"fn":1173,"description":1174,"org":1175,"tags":1176,"stars":1080,"repoUrl":1081,"updatedAt":1184},"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},[1177,1180,1181],{"name":1178,"slug":1179,"type":15},"Code Review","code-review",{"name":1103,"slug":1104,"type":15},{"name":1182,"slug":1183,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":1186,"name":1186,"fn":1187,"description":1188,"org":1189,"tags":1190,"stars":1080,"repoUrl":1081,"updatedAt":1193},"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},[1191,1192],{"name":1154,"slug":1155,"type":15},{"name":1157,"slug":1158,"type":15},"2026-07-14T05:25:54.928983",{"slug":1195,"name":1195,"fn":1196,"description":1197,"org":1198,"tags":1199,"stars":1080,"repoUrl":1081,"updatedAt":1204},"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},[1200,1203],{"name":1201,"slug":1202,"type":15},"Automation","automation",{"name":1090,"slug":1091,"type":15},"2026-07-30T05:29:03.275638",496,{"items":1207,"total":1301},[1208,1223,1233,1247,1257,1272,1287],{"slug":1209,"name":1209,"fn":1210,"description":1211,"org":1212,"tags":1213,"stars":20,"repoUrl":21,"updatedAt":1222},"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},[1214,1217,1220,1221],{"name":1215,"slug":1216,"type":15},"Data Analysis","data-analysis",{"name":1218,"slug":1219,"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":1224,"name":1224,"fn":1225,"description":1226,"org":1227,"tags":1228,"stars":20,"repoUrl":21,"updatedAt":1232},"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},[1229,1230,1231],{"name":1076,"slug":1077,"type":15},{"name":1168,"slug":1169,"type":15},{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":1234,"name":1234,"fn":1235,"description":1236,"org":1237,"tags":1238,"stars":20,"repoUrl":21,"updatedAt":1246},"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},[1239,1242,1243],{"name":1240,"slug":1241,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":1244,"slug":1245,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":1248,"name":1248,"fn":1249,"description":1250,"org":1251,"tags":1252,"stars":20,"repoUrl":21,"updatedAt":1256},"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},[1253,1254,1255],{"name":1215,"slug":1216,"type":15},{"name":9,"slug":8,"type":15},{"name":1157,"slug":1158,"type":15},"2026-07-17T05:29:03.913266",{"slug":1258,"name":1258,"fn":1259,"description":1260,"org":1261,"tags":1262,"stars":20,"repoUrl":21,"updatedAt":1271},"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},[1263,1264,1267,1268],{"name":1201,"slug":1202,"type":15},{"name":1265,"slug":1266,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":1269,"slug":1270,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":1273,"name":1273,"fn":1274,"description":1275,"org":1276,"tags":1277,"stars":20,"repoUrl":21,"updatedAt":1286},"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},[1278,1279,1282,1283],{"name":1076,"slug":1077,"type":15},{"name":1280,"slug":1281,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":1284,"slug":1285,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":1288,"name":1288,"fn":1289,"description":1290,"org":1291,"tags":1292,"stars":20,"repoUrl":21,"updatedAt":1300},"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},[1293,1294,1297],{"name":9,"slug":8,"type":15},{"name":1295,"slug":1296,"type":15},"Quantum Computing","quantum-computing",{"name":1298,"slug":1299,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305]