[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-cutedsl-kernel-integration":3,"mdc-axem1l-key":53,"related-repo-nvidia-cutedsl-kernel-integration":413,"related-org-nvidia-cutedsl-kernel-integration":423},{"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":48,"sourceUrl":51,"mdContent":52},"cutedsl-kernel-integration","integrate CuTeDSL kernels into cuDNN Frontend","Use when integrating a CuTeDSL\u002FCUTE DSL kernel into cuDNN Frontend as a frontend-only Python API, including APIBase wrappers, lazy cudnn exports, optional cutedsl dependencies, FE OSS documentation, and pytest coverage.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},"nvidia","NVIDIA","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fnvidia.png",[12,16,17,20],{"name":13,"slug":14,"type":15},"Documentation","documentation","tag",{"name":9,"slug":8,"type":15},{"name":18,"slug":19,"type":15},"Python","python",{"name":21,"slug":22,"type":15},"API Development","api-development",865,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fcudnn-frontend","2026-07-30T05:29:49.984149",null,201,[29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,8,46,47],"attention","blackwell","cuda","cuda-kernels","cuda-toolkit","deep-learning","flash-attention","fp8","gemm","gpu","grouped-gemm","hopper","mixture-of-experts","moe","mxfp8","normalization","nvfp4","sdpa","transformer",{"repoUrl":24,"stars":23,"forks":27,"topics":49,"description":50},[29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,8,46,47],"cuDNN Frontend is NVIDIA's modern, open-source entry point to the cuDNN library and a growing collection of high-performance open-source kernels.","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fcudnn-frontend\u002Ftree\u002FHEAD\u002Fskills\u002Fcutedsl-kernel-integration","---\nname: cutedsl-kernel-integration\ndescription: Use when integrating a CuTeDSL\u002FCUTE DSL kernel into cuDNN Frontend as a frontend-only Python API, including APIBase wrappers, lazy cudnn exports, optional cutedsl dependencies, FE OSS documentation, and pytest coverage.\n---\n\n# CuTeDSL Kernel Integration\n\nUse this skill to add or update a CuTeDSL frontend-only API in cuDNN Frontend. The goal is a complete integration: Python API, wrapper, exports, docs, and tests.\n\n## Before Editing\n\n1. Inspect the current repo state and avoid overwriting unrelated changes.\n2. Confirm every original source file needed for the integration is available. If a source file is missing, report that gap instead of inferring its contract from a related kernel.\n3. Record source provenance when it is available: upstream URL, local source path, commit, and which files map to public API modules versus private helpers.\n4. Classify the kernel before choosing a template:\n   - Kernel family: dense GEMM, GEMM fusion, grouped GEMM, discrete grouped GEMM, MoE, attention, sparse attention, or another frontend-only API family.\n   - Execution topology: single kernel, paired forward\u002Fbackward APIs, multi-kernel orchestrator, helper-kernel setup, distributed\u002Fruntime-coordinated execution, or internal scheduler.\n   - Public surface: class API, high-level wrapper, returned tensors, optional outputs, workspace ownership, and import\u002Fexport namespace.\n   - Internal support: source helper modules, schedulers, metadata utilities, and generated descriptors that must stay private to the package.\n   - Architecture variant: whether the public API needs transparent dispatch to an alternate CuTeDSL module for a newer GPU (for example Rubin `sm107` vs the default SM100 kernel). Keep the public class and wrapper unchanged when dispatch is internal.\n5. Read `references\u002Fintegration-pattern.md` for the detailed repo conventions before implementing.\n\n## Integration Workflow\n\n1. Add or update the operation package under the closest existing family, such as `python\u002Fcudnn\u002F\u003Coperation>\u002F`, `python\u002Fcudnn\u002Fgrouped_gemm\u002F\u003Coperation>\u002F`, `python\u002Fcudnn\u002Fdiscrete_grouped_gemm\u002F\u003Coperation>\u002F`, or `python\u002Fcudnn\u002Fsdpa\u002F\u003Cdirection>\u002F`.\n2. Implement the class API by extending `APIBase`; keep constructor descriptors, `check_support()`, `compile()`, and `execute()` consistent with the closest template.\n3. Add a high-level wrapper that allocates outputs, caches\u002Freuses compiled kernels where the template does, and returns a `TupleDict`.\n4. Export the public class and wrapper through the operation\u002Ffamily `__init__.py` files and `_LAZY_OPTIONAL_IMPORTS` in `python\u002Fcudnn\u002F__init__.py`.\n5. Reuse the existing `cutedsl` optional dependency unless the new kernel truly needs an additional package.\n6. Add FE OSS documentation and update the relevant overview or operation index links.\n7. Add tests under `test\u002Fpython\u002Ffe_api\u002F`, including support validation and numerical\u002Freference coverage when executable.\n8. For grouped\u002Fdiscrete\u002FMoE\u002FSDPA kernels, preserve the source helper and scheduler topology; shared helper modules should be internal package files, not public `cudnn` exports.\n9. When an existing SM100 kernel needs a Rubin (`sm107`) variant, follow the architecture-dispatch pattern in `references\u002Fintegration-pattern.md` instead of exposing a new public API. Current examples: `grouped_gemm_quant`, `grouped_gemm_glu`, and `grouped_gemm_dglu`.\n\n## Verification\n\n- Run focused formatting or tests for the files changed.\n- At minimum for skill-only edits, verify this `SKILL.md` has valid frontmatter and all referenced paths exist.\n- For kernel integrations, run the relevant `pytest test\u002Fpython\u002Ffe_api\u002Ftest_\u003Coperation>.py` target when the environment has the required GPU and optional dependencies; otherwise report the skipped verification explicitly.\n- For architecture-dispatch work, also run `pytest test\u002Fpython\u002Ffe_api\u002Ftest_rubin_kernel_dispatch.py`. On Rubin hardware, the existing FE API e2e tests for the affected operation should still pass without API changes.\n",{"data":54,"body":55},{"name":4,"description":6},{"type":56,"children":57},"root",[58,66,72,79,155,161,360,366],{"type":59,"tag":60,"props":61,"children":62},"element","h1",{"id":4},[63],{"type":64,"value":65},"text","CuTeDSL Kernel Integration",{"type":59,"tag":67,"props":68,"children":69},"p",{},[70],{"type":64,"value":71},"Use this skill to add or update a CuTeDSL frontend-only API in cuDNN Frontend. The goal is a complete integration: Python API, wrapper, exports, docs, and tests.",{"type":59,"tag":73,"props":74,"children":76},"h2",{"id":75},"before-editing",[77],{"type":64,"value":78},"Before Editing",{"type":59,"tag":80,"props":81,"children":82},"ol",{},[83,89,94,99,142],{"type":59,"tag":84,"props":85,"children":86},"li",{},[87],{"type":64,"value":88},"Inspect the current repo state and avoid overwriting unrelated changes.",{"type":59,"tag":84,"props":90,"children":91},{},[92],{"type":64,"value":93},"Confirm every original source file needed for the integration is available. If a source file is missing, report that gap instead of inferring its contract from a related kernel.",{"type":59,"tag":84,"props":95,"children":96},{},[97],{"type":64,"value":98},"Record source provenance when it is available: upstream URL, local source path, commit, and which files map to public API modules versus private helpers.",{"type":59,"tag":84,"props":100,"children":101},{},[102,104],{"type":64,"value":103},"Classify the kernel before choosing a template:\n",{"type":59,"tag":105,"props":106,"children":107},"ul",{},[108,113,118,123,128],{"type":59,"tag":84,"props":109,"children":110},{},[111],{"type":64,"value":112},"Kernel family: dense GEMM, GEMM fusion, grouped GEMM, discrete grouped GEMM, MoE, attention, sparse attention, or another frontend-only API family.",{"type":59,"tag":84,"props":114,"children":115},{},[116],{"type":64,"value":117},"Execution topology: single kernel, paired forward\u002Fbackward APIs, multi-kernel orchestrator, helper-kernel setup, distributed\u002Fruntime-coordinated execution, or internal scheduler.",{"type":59,"tag":84,"props":119,"children":120},{},[121],{"type":64,"value":122},"Public surface: class API, high-level wrapper, returned tensors, optional outputs, workspace ownership, and import\u002Fexport namespace.",{"type":59,"tag":84,"props":124,"children":125},{},[126],{"type":64,"value":127},"Internal support: source helper modules, schedulers, metadata utilities, and generated descriptors that must stay private to the package.",{"type":59,"tag":84,"props":129,"children":130},{},[131,133,140],{"type":64,"value":132},"Architecture variant: whether the public API needs transparent dispatch to an alternate CuTeDSL module for a newer GPU (for example Rubin ",{"type":59,"tag":134,"props":135,"children":137},"code",{"className":136},[],[138],{"type":64,"value":139},"sm107",{"type":64,"value":141}," vs the default SM100 kernel). Keep the public class and wrapper unchanged when dispatch is internal.",{"type":59,"tag":84,"props":143,"children":144},{},[145,147,153],{"type":64,"value":146},"Read ",{"type":59,"tag":134,"props":148,"children":150},{"className":149},[],[151],{"type":64,"value":152},"references\u002Fintegration-pattern.md",{"type":64,"value":154}," for the detailed repo conventions before implementing.",{"type":59,"tag":73,"props":156,"children":158},{"id":157},"integration-workflow",[159],{"type":64,"value":160},"Integration Workflow",{"type":59,"tag":80,"props":162,"children":163},{},[164,200,236,248,276,289,294,307,320],{"type":59,"tag":84,"props":165,"children":166},{},[167,169,175,177,183,184,190,192,198],{"type":64,"value":168},"Add or update the operation package under the closest existing family, such as ",{"type":59,"tag":134,"props":170,"children":172},{"className":171},[],[173],{"type":64,"value":174},"python\u002Fcudnn\u002F\u003Coperation>\u002F",{"type":64,"value":176},", ",{"type":59,"tag":134,"props":178,"children":180},{"className":179},[],[181],{"type":64,"value":182},"python\u002Fcudnn\u002Fgrouped_gemm\u002F\u003Coperation>\u002F",{"type":64,"value":176},{"type":59,"tag":134,"props":185,"children":187},{"className":186},[],[188],{"type":64,"value":189},"python\u002Fcudnn\u002Fdiscrete_grouped_gemm\u002F\u003Coperation>\u002F",{"type":64,"value":191},", or ",{"type":59,"tag":134,"props":193,"children":195},{"className":194},[],[196],{"type":64,"value":197},"python\u002Fcudnn\u002Fsdpa\u002F\u003Cdirection>\u002F",{"type":64,"value":199},".",{"type":59,"tag":84,"props":201,"children":202},{},[203,205,211,213,219,220,226,228,234],{"type":64,"value":204},"Implement the class API by extending ",{"type":59,"tag":134,"props":206,"children":208},{"className":207},[],[209],{"type":64,"value":210},"APIBase",{"type":64,"value":212},"; keep constructor descriptors, ",{"type":59,"tag":134,"props":214,"children":216},{"className":215},[],[217],{"type":64,"value":218},"check_support()",{"type":64,"value":176},{"type":59,"tag":134,"props":221,"children":223},{"className":222},[],[224],{"type":64,"value":225},"compile()",{"type":64,"value":227},", and ",{"type":59,"tag":134,"props":229,"children":231},{"className":230},[],[232],{"type":64,"value":233},"execute()",{"type":64,"value":235}," consistent with the closest template.",{"type":59,"tag":84,"props":237,"children":238},{},[239,241,247],{"type":64,"value":240},"Add a high-level wrapper that allocates outputs, caches\u002Freuses compiled kernels where the template does, and returns a ",{"type":59,"tag":134,"props":242,"children":244},{"className":243},[],[245],{"type":64,"value":246},"TupleDict",{"type":64,"value":199},{"type":59,"tag":84,"props":249,"children":250},{},[251,253,259,261,267,269,275],{"type":64,"value":252},"Export the public class and wrapper through the operation\u002Ffamily ",{"type":59,"tag":134,"props":254,"children":256},{"className":255},[],[257],{"type":64,"value":258},"__init__.py",{"type":64,"value":260}," files and ",{"type":59,"tag":134,"props":262,"children":264},{"className":263},[],[265],{"type":64,"value":266},"_LAZY_OPTIONAL_IMPORTS",{"type":64,"value":268}," in ",{"type":59,"tag":134,"props":270,"children":272},{"className":271},[],[273],{"type":64,"value":274},"python\u002Fcudnn\u002F__init__.py",{"type":64,"value":199},{"type":59,"tag":84,"props":277,"children":278},{},[279,281,287],{"type":64,"value":280},"Reuse the existing ",{"type":59,"tag":134,"props":282,"children":284},{"className":283},[],[285],{"type":64,"value":286},"cutedsl",{"type":64,"value":288}," optional dependency unless the new kernel truly needs an additional package.",{"type":59,"tag":84,"props":290,"children":291},{},[292],{"type":64,"value":293},"Add FE OSS documentation and update the relevant overview or operation index links.",{"type":59,"tag":84,"props":295,"children":296},{},[297,299,305],{"type":64,"value":298},"Add tests under ",{"type":59,"tag":134,"props":300,"children":302},{"className":301},[],[303],{"type":64,"value":304},"test\u002Fpython\u002Ffe_api\u002F",{"type":64,"value":306},", including support validation and numerical\u002Freference coverage when executable.",{"type":59,"tag":84,"props":308,"children":309},{},[310,312,318],{"type":64,"value":311},"For grouped\u002Fdiscrete\u002FMoE\u002FSDPA kernels, preserve the source helper and scheduler topology; shared helper modules should be internal package files, not public ",{"type":59,"tag":134,"props":313,"children":315},{"className":314},[],[316],{"type":64,"value":317},"cudnn",{"type":64,"value":319}," exports.",{"type":59,"tag":84,"props":321,"children":322},{},[323,325,330,332,337,339,345,346,352,353,359],{"type":64,"value":324},"When an existing SM100 kernel needs a Rubin (",{"type":59,"tag":134,"props":326,"children":328},{"className":327},[],[329],{"type":64,"value":139},{"type":64,"value":331},") variant, follow the architecture-dispatch pattern in ",{"type":59,"tag":134,"props":333,"children":335},{"className":334},[],[336],{"type":64,"value":152},{"type":64,"value":338}," instead of exposing a new public API. Current examples: ",{"type":59,"tag":134,"props":340,"children":342},{"className":341},[],[343],{"type":64,"value":344},"grouped_gemm_quant",{"type":64,"value":176},{"type":59,"tag":134,"props":347,"children":349},{"className":348},[],[350],{"type":64,"value":351},"grouped_gemm_glu",{"type":64,"value":227},{"type":59,"tag":134,"props":354,"children":356},{"className":355},[],[357],{"type":64,"value":358},"grouped_gemm_dglu",{"type":64,"value":199},{"type":59,"tag":73,"props":361,"children":363},{"id":362},"verification",[364],{"type":64,"value":365},"Verification",{"type":59,"tag":105,"props":367,"children":368},{},[369,374,387,400],{"type":59,"tag":84,"props":370,"children":371},{},[372],{"type":64,"value":373},"Run focused formatting or tests for the files changed.",{"type":59,"tag":84,"props":375,"children":376},{},[377,379,385],{"type":64,"value":378},"At minimum for skill-only edits, verify this ",{"type":59,"tag":134,"props":380,"children":382},{"className":381},[],[383],{"type":64,"value":384},"SKILL.md",{"type":64,"value":386}," has valid frontmatter and all referenced paths exist.",{"type":59,"tag":84,"props":388,"children":389},{},[390,392,398],{"type":64,"value":391},"For kernel integrations, run the relevant ",{"type":59,"tag":134,"props":393,"children":395},{"className":394},[],[396],{"type":64,"value":397},"pytest test\u002Fpython\u002Ffe_api\u002Ftest_\u003Coperation>.py",{"type":64,"value":399}," target when the environment has the required GPU and optional dependencies; otherwise report the skipped verification explicitly.",{"type":59,"tag":84,"props":401,"children":402},{},[403,405,411],{"type":64,"value":404},"For architecture-dispatch work, also run ",{"type":59,"tag":134,"props":406,"children":408},{"className":407},[],[409],{"type":64,"value":410},"pytest test\u002Fpython\u002Ffe_api\u002Ftest_rubin_kernel_dispatch.py",{"type":64,"value":412},". On Rubin hardware, the existing FE API e2e tests for the affected operation should still pass without API changes.",{"items":414,"total":422},[415],{"slug":4,"name":4,"fn":5,"description":6,"org":416,"tags":417,"stars":23,"repoUrl":24,"updatedAt":25},{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[418,419,420,421],{"name":21,"slug":22,"type":15},{"name":13,"slug":14,"type":15},{"name":9,"slug":8,"type":15},{"name":18,"slug":19,"type":15},1,{"items":424,"total":579},[425,441,457,468,480,494,507,521,534,545,559,568],{"slug":426,"name":426,"fn":427,"description":428,"org":429,"tags":430,"stars":438,"repoUrl":439,"updatedAt":440},"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},[431,432,435],{"name":13,"slug":14,"type":15},{"name":433,"slug":434,"type":15},"MCP","mcp",{"name":436,"slug":437,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":442,"name":442,"fn":443,"description":444,"org":445,"tags":446,"stars":454,"repoUrl":455,"updatedAt":456},"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},[447,450,453],{"name":448,"slug":449,"type":15},"Containers","containers",{"name":451,"slug":452,"type":15},"Deployment","deployment",{"name":18,"slug":19,"type":15},17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":458,"name":458,"fn":459,"description":460,"org":461,"tags":462,"stars":454,"repoUrl":455,"updatedAt":467},"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},[463,466],{"name":464,"slug":465,"type":15},"CI\u002FCD","ci-cd",{"name":451,"slug":452,"type":15},"2026-07-14T05:25:59.97109",{"slug":469,"name":469,"fn":470,"description":471,"org":472,"tags":473,"stars":454,"repoUrl":455,"updatedAt":479},"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},[474,475,476],{"name":464,"slug":465,"type":15},{"name":451,"slug":452,"type":15},{"name":477,"slug":478,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":481,"name":481,"fn":482,"description":483,"org":484,"tags":485,"stars":454,"repoUrl":455,"updatedAt":493},"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},[486,489,490],{"name":487,"slug":488,"type":15},"Debugging","debugging",{"name":477,"slug":478,"type":15},{"name":491,"slug":492,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":495,"name":495,"fn":496,"description":497,"org":498,"tags":499,"stars":454,"repoUrl":455,"updatedAt":506},"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},[500,503],{"name":501,"slug":502,"type":15},"Best Practices","best-practices",{"name":504,"slug":505,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":508,"name":508,"fn":509,"description":510,"org":511,"tags":512,"stars":454,"repoUrl":455,"updatedAt":520},"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},[513,516,519],{"name":514,"slug":515,"type":15},"Machine Learning","machine-learning",{"name":517,"slug":518,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":522,"name":522,"fn":523,"description":524,"org":525,"tags":526,"stars":454,"repoUrl":455,"updatedAt":533},"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},[527,530],{"name":528,"slug":529,"type":15},"QA","qa",{"name":531,"slug":532,"type":15},"Testing","testing","2026-07-14T05:25:53.673039",{"slug":535,"name":535,"fn":536,"description":537,"org":538,"tags":539,"stars":454,"repoUrl":455,"updatedAt":544},"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},[540,541],{"name":451,"slug":452,"type":15},{"name":542,"slug":543,"type":15},"Infrastructure","infrastructure","2026-07-14T05:25:49.362534",{"slug":546,"name":546,"fn":547,"description":548,"org":549,"tags":550,"stars":454,"repoUrl":455,"updatedAt":558},"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},[551,554,555],{"name":552,"slug":553,"type":15},"Code Review","code-review",{"name":477,"slug":478,"type":15},{"name":556,"slug":557,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":560,"name":560,"fn":561,"description":562,"org":563,"tags":564,"stars":454,"repoUrl":455,"updatedAt":567},"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},[565,566],{"name":528,"slug":529,"type":15},{"name":531,"slug":532,"type":15},"2026-07-14T05:25:54.928983",{"slug":569,"name":569,"fn":570,"description":571,"org":572,"tags":573,"stars":454,"repoUrl":455,"updatedAt":578},"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},[574,577],{"name":575,"slug":576,"type":15},"Automation","automation",{"name":464,"slug":465,"type":15},"2026-07-30T05:29:03.275638",496]