[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-dgx-station-inference":3,"mdc--poxc4m-key":34,"related-repo-nvidia-dgx-station-inference":389,"related-org-nvidia-dgx-station-inference":497},{"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},"dgx-station-inference","launch and tune model inference on DGX","Resolve, tune, preflight, launch, verify, inspect, and stop exact-model inference on NVIDIA DGX Station through dgx-assist. Use for vLLM or SGLang container selection, NGC versus upstream, GPU memory utilization, CPU or KV offload, HBM fit, KV-cache sizing, ISL or context length, prefix caching, chunked prefill, batching, concurrency, performance tuning, serving or deploying a named model, an OpenAI-compatible endpoint, Station recipe models, or an owned inference service. Require an exact model ID for recipe resolution or model-specific tuning, and never recommend or substitute a different model.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},"nvidia","NVIDIA","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fnvidia.png",[12,16,19,22],{"name":13,"slug":14,"type":15},"Performance","performance","tag",{"name":17,"slug":18,"type":15},"LLM","llm",{"name":20,"slug":21,"type":15},"Deployment","deployment",{"name":9,"slug":8,"type":15},1144,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fdgx-spark-playbooks","2026-08-05T05:58:30.813354",null,249,[],{"repoUrl":24,"stars":23,"forks":27,"topics":30,"description":31},[],"Collection of step-by-step playbooks for setting up AI\u002FML workloads on NVIDIA DGX Spark devices with Blackwell architecture.","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fdgx-spark-playbooks\u002Ftree\u002FHEAD\u002Fnvidia\u002Fstation-ai-skills\u002Fassets\u002Fskills\u002Fdgx-station-inference","---\nname: dgx-station-inference\ndescription: Resolve, tune, preflight, launch, verify, inspect, and stop exact-model inference on NVIDIA DGX Station through dgx-assist. Use for vLLM or SGLang container selection, NGC versus upstream, GPU memory utilization, CPU or KV offload, HBM fit, KV-cache sizing, ISL or context length, prefix caching, chunked prefill, batching, concurrency, performance tuning, serving or deploying a named model, an OpenAI-compatible endpoint, Station recipe models, or an owned inference service. Require an exact model ID for recipe resolution or model-specific tuning, and never recommend or substitute a different model.\n---\n\n# DGX Station inference\n\nServe only trusted, published, current, physically validated recipes. Bundled\nrecipes are bound to the installed package; downloaded refreshes must be\nsignature-verified.\n\n## Workflow\n\n1. Require the user's exact model ID. If absent, run `scripts\u002Fdgx-assist recipe models --json`, show the alphabetical IDs and runnable states without ranking them, then ask which exact model to use.\n2. Run `scripts\u002Fdgx-assist system inspect --json`.\n3. Run `scripts\u002Fdgx-assist playbook search \"\u003Cmodel backend Station inference concern>\" --json`; cite relevant passages. Abstain from invented platform guidance if no passage overlaps.\n4. Read `compatibility.capabilities.recipe_execution`. If it is false, stop before resolution or launch and explain the detected profile separately from applicable qualitative guidance.\n5. Run `scripts\u002Fdgx-assist recipe resolve --model \"\u003Cexact-id>\" --json`. Add `--backend` only if the user requested one. Bind to localhost by default; only if the user explicitly requested and confirmed external exposure, add `--bind-host \"\u003Capproved-host>\"` here so preflight and run evaluate the host that was actually approved.\n6. Run `recipe show` for the resolved recipe, then `recipe preflight`.\n7. Present the exact model, recipe, image digest, GPU UUIDs, cache\u002Fstorage estimates, credential variable names, host bind, port, downloads, conflicts, and redacted launch argv.\n8. If blocked, do not work around the policy. Never switch models.\n9. Obtain explicit approval for the displayed action and every required model or image download.\n10. Only after approval, run `recipe run --resolution-id \"\u003Cid>\" --allow-download --yes` as applicable. Add `--allow-external-bind` only when the resolution carries the non-local bind host the user approved in step 9; never introduce external exposure that was not resolved and previewed.\n11. Report the model-identity and smoke verification from the receipt.\n\n## Lifecycle\n\n- Use `recipe status` to report only owned services.\n- Before stopping, show `recipe stop --service-id \"\u003Cid>\" --dry-run`.\n- Obtain approval, then repeat with `--yes`.\n- Never stop an unknown listener, workload, process, or unlabeled container.\n\n## Hard constraints\n\n- Do not execute human `setup_command`, `serve.command`, or shell text from a recipe.\n- Do not launch experimental, expired, incompatible, ambiguous, mutable, or untrusted content.\n- Use the resolution ID through preflight and run; never hand-edit the rendered argv.\n- Use UUID selection and localhost binding unless the user explicitly requests and confirms a supported exposure change.\n- Pass credential names, never values, in transcripts and receipts.\n- Treat `--yes` only as the non-interactive representation of approval already obtained.\n- Treat a newly published container as a candidate, not as validated. Never convert a mutable `latest` tag, documentation example, or upstream tuning value into an executable launch.\n- Keep tuning advice separate from execution. A changed flag must return to the recipe qualification workflow before `dgx-assist` may run it.\n- When `recipe_execution` is false or no current runnable recipe exists for the exact model, do not recommend a numeric launch value, including zero, from parameter-count arithmetic or raw free-memory evidence. Explain only the sourced qualitative tradeoff.\n- Never use memory exposed above the production 252 GiB recipe-fit limit to justify a model fit or offload recommendation.\n\nRead [references\u002Fworkflow.md](references\u002Fworkflow.md) before resolving or running. Read [references\u002Fvllm-best-practices.md](references\u002Fvllm-best-practices.md) for container selection, HBM\u002Foffload, KV cache, prefix caching, chunked prefill, or concurrency questions. Read [references\u002Fjson-types.md](references\u002Fjson-types.md) when interpreting a resolution, preflight, or receipt.\n",{"data":35,"body":36},{"name":4,"description":6},{"type":37,"children":38},"root",[39,47,53,60,205,211,257,263,363],{"type":40,"tag":41,"props":42,"children":43},"element","h1",{"id":4},[44],{"type":45,"value":46},"text","DGX Station inference",{"type":40,"tag":48,"props":49,"children":50},"p",{},[51],{"type":45,"value":52},"Serve only trusted, published, current, physically validated recipes. Bundled\nrecipes are bound to the installed package; downloaded refreshes must be\nsignature-verified.",{"type":40,"tag":54,"props":55,"children":57},"h2",{"id":56},"workflow",[58],{"type":45,"value":59},"Workflow",{"type":40,"tag":61,"props":62,"children":63},"ol",{},[64,79,92,104,117,145,164,169,174,179,200],{"type":40,"tag":65,"props":66,"children":67},"li",{},[68,70,77],{"type":45,"value":69},"Require the user's exact model ID. If absent, run ",{"type":40,"tag":71,"props":72,"children":74},"code",{"className":73},[],[75],{"type":45,"value":76},"scripts\u002Fdgx-assist recipe models --json",{"type":45,"value":78},", show the alphabetical IDs and runnable states without ranking them, then ask which exact model to use.",{"type":40,"tag":65,"props":80,"children":81},{},[82,84,90],{"type":45,"value":83},"Run ",{"type":40,"tag":71,"props":85,"children":87},{"className":86},[],[88],{"type":45,"value":89},"scripts\u002Fdgx-assist system inspect --json",{"type":45,"value":91},".",{"type":40,"tag":65,"props":93,"children":94},{},[95,96,102],{"type":45,"value":83},{"type":40,"tag":71,"props":97,"children":99},{"className":98},[],[100],{"type":45,"value":101},"scripts\u002Fdgx-assist playbook search \"\u003Cmodel backend Station inference concern>\" --json",{"type":45,"value":103},"; cite relevant passages. Abstain from invented platform guidance if no passage overlaps.",{"type":40,"tag":65,"props":105,"children":106},{},[107,109,115],{"type":45,"value":108},"Read ",{"type":40,"tag":71,"props":110,"children":112},{"className":111},[],[113],{"type":45,"value":114},"compatibility.capabilities.recipe_execution",{"type":45,"value":116},". If it is false, stop before resolution or launch and explain the detected profile separately from applicable qualitative guidance.",{"type":40,"tag":65,"props":118,"children":119},{},[120,121,127,129,135,137,143],{"type":45,"value":83},{"type":40,"tag":71,"props":122,"children":124},{"className":123},[],[125],{"type":45,"value":126},"scripts\u002Fdgx-assist recipe resolve --model \"\u003Cexact-id>\" --json",{"type":45,"value":128},". Add ",{"type":40,"tag":71,"props":130,"children":132},{"className":131},[],[133],{"type":45,"value":134},"--backend",{"type":45,"value":136}," only if the user requested one. Bind to localhost by default; only if the user explicitly requested and confirmed external exposure, add ",{"type":40,"tag":71,"props":138,"children":140},{"className":139},[],[141],{"type":45,"value":142},"--bind-host \"\u003Capproved-host>\"",{"type":45,"value":144}," here so preflight and run evaluate the host that was actually approved.",{"type":40,"tag":65,"props":146,"children":147},{},[148,149,155,157,163],{"type":45,"value":83},{"type":40,"tag":71,"props":150,"children":152},{"className":151},[],[153],{"type":45,"value":154},"recipe show",{"type":45,"value":156}," for the resolved recipe, then ",{"type":40,"tag":71,"props":158,"children":160},{"className":159},[],[161],{"type":45,"value":162},"recipe preflight",{"type":45,"value":91},{"type":40,"tag":65,"props":165,"children":166},{},[167],{"type":45,"value":168},"Present the exact model, recipe, image digest, GPU UUIDs, cache\u002Fstorage estimates, credential variable names, host bind, port, downloads, conflicts, and redacted launch argv.",{"type":40,"tag":65,"props":170,"children":171},{},[172],{"type":45,"value":173},"If blocked, do not work around the policy. Never switch models.",{"type":40,"tag":65,"props":175,"children":176},{},[177],{"type":45,"value":178},"Obtain explicit approval for the displayed action and every required model or image download.",{"type":40,"tag":65,"props":180,"children":181},{},[182,184,190,192,198],{"type":45,"value":183},"Only after approval, run ",{"type":40,"tag":71,"props":185,"children":187},{"className":186},[],[188],{"type":45,"value":189},"recipe run --resolution-id \"\u003Cid>\" --allow-download --yes",{"type":45,"value":191}," as applicable. Add ",{"type":40,"tag":71,"props":193,"children":195},{"className":194},[],[196],{"type":45,"value":197},"--allow-external-bind",{"type":45,"value":199}," only when the resolution carries the non-local bind host the user approved in step 9; never introduce external exposure that was not resolved and previewed.",{"type":40,"tag":65,"props":201,"children":202},{},[203],{"type":45,"value":204},"Report the model-identity and smoke verification from the receipt.",{"type":40,"tag":54,"props":206,"children":208},{"id":207},"lifecycle",[209],{"type":45,"value":210},"Lifecycle",{"type":40,"tag":212,"props":213,"children":214},"ul",{},[215,228,240,252],{"type":40,"tag":65,"props":216,"children":217},{},[218,220,226],{"type":45,"value":219},"Use ",{"type":40,"tag":71,"props":221,"children":223},{"className":222},[],[224],{"type":45,"value":225},"recipe status",{"type":45,"value":227}," to report only owned services.",{"type":40,"tag":65,"props":229,"children":230},{},[231,233,239],{"type":45,"value":232},"Before stopping, show ",{"type":40,"tag":71,"props":234,"children":236},{"className":235},[],[237],{"type":45,"value":238},"recipe stop --service-id \"\u003Cid>\" --dry-run",{"type":45,"value":91},{"type":40,"tag":65,"props":241,"children":242},{},[243,245,251],{"type":45,"value":244},"Obtain approval, then repeat with ",{"type":40,"tag":71,"props":246,"children":248},{"className":247},[],[249],{"type":45,"value":250},"--yes",{"type":45,"value":91},{"type":40,"tag":65,"props":253,"children":254},{},[255],{"type":45,"value":256},"Never stop an unknown listener, workload, process, or unlabeled container.",{"type":40,"tag":54,"props":258,"children":260},{"id":259},"hard-constraints",[261],{"type":45,"value":262},"Hard constraints",{"type":40,"tag":212,"props":264,"children":265},{},[266,287,292,297,302,307,319,332,345,358],{"type":40,"tag":65,"props":267,"children":268},{},[269,271,277,279,285],{"type":45,"value":270},"Do not execute human ",{"type":40,"tag":71,"props":272,"children":274},{"className":273},[],[275],{"type":45,"value":276},"setup_command",{"type":45,"value":278},", ",{"type":40,"tag":71,"props":280,"children":282},{"className":281},[],[283],{"type":45,"value":284},"serve.command",{"type":45,"value":286},", or shell text from a recipe.",{"type":40,"tag":65,"props":288,"children":289},{},[290],{"type":45,"value":291},"Do not launch experimental, expired, incompatible, ambiguous, mutable, or untrusted content.",{"type":40,"tag":65,"props":293,"children":294},{},[295],{"type":45,"value":296},"Use the resolution ID through preflight and run; never hand-edit the rendered argv.",{"type":40,"tag":65,"props":298,"children":299},{},[300],{"type":45,"value":301},"Use UUID selection and localhost binding unless the user explicitly requests and confirms a supported exposure change.",{"type":40,"tag":65,"props":303,"children":304},{},[305],{"type":45,"value":306},"Pass credential names, never values, in transcripts and receipts.",{"type":40,"tag":65,"props":308,"children":309},{},[310,312,317],{"type":45,"value":311},"Treat ",{"type":40,"tag":71,"props":313,"children":315},{"className":314},[],[316],{"type":45,"value":250},{"type":45,"value":318}," only as the non-interactive representation of approval already obtained.",{"type":40,"tag":65,"props":320,"children":321},{},[322,324,330],{"type":45,"value":323},"Treat a newly published container as a candidate, not as validated. Never convert a mutable ",{"type":40,"tag":71,"props":325,"children":327},{"className":326},[],[328],{"type":45,"value":329},"latest",{"type":45,"value":331}," tag, documentation example, or upstream tuning value into an executable launch.",{"type":40,"tag":65,"props":333,"children":334},{},[335,337,343],{"type":45,"value":336},"Keep tuning advice separate from execution. A changed flag must return to the recipe qualification workflow before ",{"type":40,"tag":71,"props":338,"children":340},{"className":339},[],[341],{"type":45,"value":342},"dgx-assist",{"type":45,"value":344}," may run it.",{"type":40,"tag":65,"props":346,"children":347},{},[348,350,356],{"type":45,"value":349},"When ",{"type":40,"tag":71,"props":351,"children":353},{"className":352},[],[354],{"type":45,"value":355},"recipe_execution",{"type":45,"value":357}," is false or no current runnable recipe exists for the exact model, do not recommend a numeric launch value, including zero, from parameter-count arithmetic or raw free-memory evidence. Explain only the sourced qualitative tradeoff.",{"type":40,"tag":65,"props":359,"children":360},{},[361],{"type":45,"value":362},"Never use memory exposed above the production 252 GiB recipe-fit limit to justify a model fit or offload recommendation.",{"type":40,"tag":48,"props":364,"children":365},{},[366,367,373,375,380,382,387],{"type":45,"value":108},{"type":40,"tag":368,"props":369,"children":371},"a",{"href":370},"references\u002Fworkflow.md",[372],{"type":45,"value":370},{"type":45,"value":374}," before resolving or running. Read ",{"type":40,"tag":368,"props":376,"children":378},{"href":377},"references\u002Fvllm-best-practices.md",[379],{"type":45,"value":377},{"type":45,"value":381}," for container selection, HBM\u002Foffload, KV cache, prefix caching, chunked prefill, or concurrency questions. Read ",{"type":40,"tag":368,"props":383,"children":385},{"href":384},"references\u002Fjson-types.md",[386],{"type":45,"value":384},{"type":45,"value":388}," when interpreting a resolution, preflight, or receipt.",{"items":390,"total":496},[391,411,424,439,454,465,482],{"slug":392,"name":392,"fn":393,"description":394,"org":395,"tags":396,"stars":23,"repoUrl":24,"updatedAt":410},"analysis-methods","write Python analysis code for FHIR data","Teaches the analyst agent how to write correct, robust Python analysis code for FHIR clinical data using pandas, matplotlib, and scipy.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[397,400,403,406,407],{"name":398,"slug":399,"type":15},"Data Analysis","data-analysis",{"name":401,"slug":402,"type":15},"FHIR","fhir",{"name":404,"slug":405,"type":15},"Healthcare","healthcare",{"name":9,"slug":8,"type":15},{"name":408,"slug":409,"type":15},"Python","python","2026-07-14T05:35:59.037962",{"slug":412,"name":412,"fn":413,"description":414,"org":415,"tags":416,"stars":23,"repoUrl":24,"updatedAt":423},"case-summary","summarize clinical patient cases from FHIR","Prepare a complete clinical case summary for a patient from FHIR endpoints. Use when asked to summarize a patient, compile a case, or prepare for tumor board.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[417,418,419,420],{"name":401,"slug":402,"type":15},{"name":404,"slug":405,"type":15},{"name":9,"slug":8,"type":15},{"name":421,"slug":422,"type":15},"Summarization","summarization","2026-07-14T05:35:52.790528",{"slug":425,"name":425,"fn":426,"description":427,"org":428,"tags":429,"stars":23,"repoUrl":24,"updatedAt":438},"clinical-delegation","delegate clinical tasks to specialist agents","How to delegate clinical tasks to specialist agents. Always use sub-agent runtime with explicit agentId — never ACP. Never call FHIR via web_fetch.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[430,433,434,437],{"name":431,"slug":432,"type":15},"Agents","agents",{"name":404,"slug":405,"type":15},{"name":435,"slug":436,"type":15},"Multi-Agent","multi-agent",{"name":9,"slug":8,"type":15},"2026-07-14T05:35:55.294972",{"slug":440,"name":440,"fn":441,"description":442,"org":443,"tags":444,"stars":23,"repoUrl":24,"updatedAt":453},"clinical-knowledge","provide clinical reference and regulatory context","Teaches agents clinical reference ranges, condition codes, quality measure definitions, drug classifications, and regulatory context so they can flag abnormal values and identify care gaps.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[445,448,449,450],{"name":446,"slug":447,"type":15},"Clinical Trials","clinical-trials",{"name":404,"slug":405,"type":15},{"name":9,"slug":8,"type":15},{"name":451,"slug":452,"type":15},"Regulatory Compliance","regulatory-compliance","2026-07-14T05:35:56.550833",{"slug":455,"name":455,"fn":456,"description":457,"org":458,"tags":459,"stars":23,"repoUrl":24,"updatedAt":464},"cohort-compare","analyze patient cohorts from FHIR endpoints","Analyze a cohort of patients from FHIR endpoints to find care gaps and patterns. Use when asked to compare patients, find quality gaps, or analyze a population.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[460,461,462,463],{"name":398,"slug":399,"type":15},{"name":401,"slug":402,"type":15},{"name":404,"slug":405,"type":15},{"name":9,"slug":8,"type":15},"2026-07-14T05:35:51.543095",{"slug":466,"name":466,"fn":467,"description":468,"org":469,"tags":470,"stars":23,"repoUrl":24,"updatedAt":481},"dgx-diagnose","diagnose NVIDIA DGX Station hardware issues","Diagnose common DGX Station GB300 issues — CUDA crashes, wrong-GPU targeting, vLLM\u002FSGLang container bugs, MIG state problems, NVLink\u002FFabric Manager errors, X\u002FVulkan failures, HuggingFace auth, and port conflicts. Use when the user reports a GPU error, inference server crash, MIG problem, or any unexplained DGX Station failure.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[471,474,477,478],{"name":472,"slug":473,"type":15},"AI Infrastructure","ai-infrastructure",{"name":475,"slug":476,"type":15},"Debugging","debugging",{"name":9,"slug":8,"type":15},{"name":479,"slug":480,"type":15},"Observability","observability","2026-07-14T05:31:04.085598",{"slug":483,"name":483,"fn":484,"description":485,"org":486,"tags":487,"stars":23,"repoUrl":24,"updatedAt":495},"dgx-station","manage NVIDIA DGX Station development","Inspect and guide NVIDIA DGX Station GB300 development using the local dgx-assist CLI and pinned NVIDIA playbooks. Use for general Station platform questions, Software 1.0 or 2.0 compatibility, GB300 or RTX GPU selection, UUID ordering, mixed ATS\u002FHMM coherency, CDMM, general containers, CDI, CUDA visibility, or vsloshd power-sloshing behavior. Do not use for vLLM or SGLang container selection or tuning, serving a named model, changing MIG, or troubleshooting a reported failure when the dedicated Station skill applies.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[488,491,492],{"name":489,"slug":490,"type":15},"CLI","cli",{"name":9,"slug":8,"type":15},{"name":493,"slug":494,"type":15},"Operations","operations","2026-08-05T05:58:16.736667",15,{"items":498,"total":651},[499,517,531,542,554,566,579,593,606,617,631,640],{"slug":500,"name":500,"fn":501,"description":502,"org":503,"tags":504,"stars":514,"repoUrl":515,"updatedAt":516},"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},[505,508,511],{"name":506,"slug":507,"type":15},"Documentation","documentation",{"name":509,"slug":510,"type":15},"MCP","mcp",{"name":512,"slug":513,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":518,"name":518,"fn":519,"description":520,"org":521,"tags":522,"stars":528,"repoUrl":529,"updatedAt":530},"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},[523,526,527],{"name":524,"slug":525,"type":15},"Containers","containers",{"name":20,"slug":21,"type":15},{"name":408,"slug":409,"type":15},17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":532,"name":532,"fn":533,"description":534,"org":535,"tags":536,"stars":528,"repoUrl":529,"updatedAt":541},"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},[537,540],{"name":538,"slug":539,"type":15},"CI\u002FCD","ci-cd",{"name":20,"slug":21,"type":15},"2026-07-14T05:25:59.97109",{"slug":543,"name":543,"fn":544,"description":545,"org":546,"tags":547,"stars":528,"repoUrl":529,"updatedAt":553},"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},[548,549,550],{"name":538,"slug":539,"type":15},{"name":20,"slug":21,"type":15},{"name":551,"slug":552,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":555,"name":555,"fn":556,"description":557,"org":558,"tags":559,"stars":528,"repoUrl":529,"updatedAt":565},"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},[560,561,562],{"name":475,"slug":476,"type":15},{"name":551,"slug":552,"type":15},{"name":563,"slug":564,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":567,"name":567,"fn":568,"description":569,"org":570,"tags":571,"stars":528,"repoUrl":529,"updatedAt":578},"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},[572,575],{"name":573,"slug":574,"type":15},"Best Practices","best-practices",{"name":576,"slug":577,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":580,"name":580,"fn":581,"description":582,"org":583,"tags":584,"stars":528,"repoUrl":529,"updatedAt":592},"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},[585,588,591],{"name":586,"slug":587,"type":15},"Machine Learning","machine-learning",{"name":589,"slug":590,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":594,"name":594,"fn":595,"description":596,"org":597,"tags":598,"stars":528,"repoUrl":529,"updatedAt":605},"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},[599,602],{"name":600,"slug":601,"type":15},"QA","qa",{"name":603,"slug":604,"type":15},"Testing","testing","2026-07-14T05:25:53.673039",{"slug":607,"name":607,"fn":608,"description":609,"org":610,"tags":611,"stars":528,"repoUrl":529,"updatedAt":616},"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},[612,613],{"name":20,"slug":21,"type":15},{"name":614,"slug":615,"type":15},"Infrastructure","infrastructure","2026-07-14T05:25:49.362534",{"slug":618,"name":618,"fn":619,"description":620,"org":621,"tags":622,"stars":528,"repoUrl":529,"updatedAt":630},"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},[623,626,627],{"name":624,"slug":625,"type":15},"Code Review","code-review",{"name":551,"slug":552,"type":15},{"name":628,"slug":629,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":632,"name":632,"fn":633,"description":634,"org":635,"tags":636,"stars":528,"repoUrl":529,"updatedAt":639},"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},[637,638],{"name":600,"slug":601,"type":15},{"name":603,"slug":604,"type":15},"2026-07-14T05:25:54.928983",{"slug":641,"name":641,"fn":642,"description":643,"org":644,"tags":645,"stars":528,"repoUrl":529,"updatedAt":650},"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},[646,649],{"name":647,"slug":648,"type":15},"Automation","automation",{"name":538,"slug":539,"type":15},"2026-07-30T05:29:03.275638",525]