[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-nemotron-retrieval-recipes":3,"mdc--h7bbtv-key":34,"related-org-nvidia-nemotron-retrieval-recipes":1103,"related-repo-nvidia-nemotron-retrieval-recipes":1260},{"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},"nemotron-retrieval-recipes","manage Nemotron retrieval recipes","Use when planning, debugging, tuning, evaluating, exporting, or deploying public Nemotron `embed`\u002F`rerank` retrieval recipes.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},"nvidia","NVIDIA","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fnvidia.png",[12,16,19,20],{"name":13,"slug":14,"type":15},"LLM","llm","tag",{"name":17,"slug":18,"type":15},"Deployment","deployment",{"name":9,"slug":8,"type":15},{"name":21,"slug":22,"type":15},"Search","search",2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-07-17T05:28:58.911896","Apache-2.0",281,[],{"repoUrl":24,"stars":23,"forks":27,"topics":30,"description":31},[],"AI agent skills published by NVIDIA","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills\u002Ftree\u002FHEAD\u002Fskills\u002Fnemotron-retrieval-recipes","---\nname: nemotron-retrieval-recipes\nversion: \"0.2.0\"\nauthor: \"NVIDIA Nemotron Team \u003Cnoreply@nvidia.com>\"\nlicense: Apache-2.0\ntags:\n  - nemotron\n  - retrieval\n  - fine-tuning\n  - embeddings\n  - reranking\nmetadata:\n  author: \"NVIDIA Nemotron Team \u003Cnoreply@nvidia.com>\"\n  tags:\n    - nemotron\n    - retrieval\n    - fine-tuning\n    - embeddings\n    - reranking\ntools:\n  - Read\n  - Bash\n  - Search\ndescription: Use when planning, debugging, tuning, evaluating, exporting, or deploying public Nemotron `embed`\u002F`rerank` retrieval recipes.\n---\n\n# Nemotron Retrieval Recipes\n\nInvocation: `$nemotron-retrieval-recipes`.\n\n## Purpose\n\nUse this skill to work with public Nemotron embedding and reranking retrieval recipes in a source checkout or installed package. Prefer the current checkout over memory, because the recipe CLI, configs, containers, and output paths are actively changing. Treat each recipe family as available only after its recipe directory and matching CLI files are present.\n\nThis is a public product skill, not contributor-only guidance. Its value over static docs is to make an agent route the user's retrieval failure to the right recipe family, reconcile docs with the current checkout, avoid accidental long-running launches, preserve secrets, and return concrete preview\u002Fexecution\u002Frun-report commands.\n\nUse it only for tasks tied to the public Nemotron `embed` or `rerank` recipe flow. If the request is unrelated retrieval theory, generic vector database selection, generic benchmark advice, or non-recipe Docker\u002FSlurm\u002FNIM troubleshooting, stop with a short scope note and do not inspect recipe files in that turn.\n\n## Security Notes\n\nUse `Bash` for repo-scoped inspection, help, dry-run, and user-approved execution commands. Do not run API, GPU, Docker, Slurm, NIM, or other long-running work unless the user explicitly asks for it. Before Stage 0 SDG for either family, confirm the user's data-governance policy permits sending corpus content to the configured inference endpoints; otherwise use an approved private or air-gapped path. Never run broad environment dumps or commands that expose secret values. Prefer dotlist overrides and config review over editing recipe defaults.\n\n## Source Priority\n\nResolve conflicts in this order:\n\n1. Current checkout recipe, CLI, config, and source files.\n2. Bundled references in this skill.\n3. User-provided docs or saved snippets.\n4. Memory.\n\nFor runnable commands, treat the current checkout as authoritative. If a required recipe directory, CLI command, config, or env profile is missing, report the blocker instead of guessing.\n\n## Prerequisites\n\n- Repo environment: `uv sync --all-extras` or the smallest relevant extra documented by the checkout.\n- Stage 0 SDG: `NVIDIA_API_KEY`; never ask users to paste secret values.\n- Stages 1–3 GPU work: CUDA\u002FNVIDIA driver availability and enough VRAM.\n- Stage 4 export: NeMo Export-Deploy container when using TensorRT. The default Nemotron 3 Embed profile intentionally skips export.\n- Stage 5 deploy: Docker. Default Nemotron 3 Embed can use the checked-in vLLM path with `backend=vllm`, or a compatible `NEMOTRON3_EMBED_NIM_IMAGE` with `backend=nim`; Llama Embed and rerank deployment may require NGC access and `NGC_API_KEY`.\n- Remote execution: root `env.toml` profile for `--run` or `--batch`; load `references\u002Fremote.md` when remote scheduling, logs, or GPU placement matter.\n\n## Instructions\n\n1. Identify the recipe family.\n   - Use `references\u002Fembed.md` for embedding, embed, bi-encoder, vector search, first-stage retrieval, low Recall@k, missing relevant documents, NIM embeddings, or `nemotron embed`.\n   - Use `references\u002Frerank.md` for rerank, reranker, cross-encoder, second-stage retrieval, acceptable recall but poor top-rank ordering, low nDCG with good Recall, or `nemotron rerank`.\n   - Use both references only when the user asks about both families or asks which family to choose.\n2. For `embed`, choose one model profile before composing stage commands.\n   - Run `uv run nemotron embed info` when the requested model is unclear.\n   - Use `-c default` for `nvidia\u002FNemotron-3-Embed-1B-BF16`.\n   - Use `-c llama` for `nvidia\u002Fllama-nemotron-embed-1b-v2` and its export path.\n   - Carry the selected profile and `artifact_root` through every stage; never combine artifacts from the two profiles.\n3. Choose the model family to tune from the retrieval failure mode.\n   - Prefer embedding fine-tuning when relevant documents are absent from the candidate set.\n   - Prefer reranker fine-tuning when relevant documents are retrieved but ordered poorly near the top.\n   - For production retrieval stacks, remember that these are complementary: embed first, rerank candidates second.\n4. Identify the intent: plan a run, execute a stage, debug a failure, tune hyperparameters, interpret metrics, export\u002Fdeploy a model, inspect configs, or propose dotlist overrides.\n5. Inspect the current public surface before acting:\n   - Recipe files: `src\u002Fnemotron\u002Frecipes\u002F\u003Cembed|rerank>\u002F`\n   - CLI files: `src\u002Fnemotron\u002Fcli\u002Fcommands\u002F\u003Cembed|rerank>\u002F`\n   - Configs: `src\u002Fnemotron\u002Frecipes\u002F\u003Cfamily>\u002Fstage*\u002Fconfig\u002F\u003Cprofile>.yaml`\n   - Help and dry runs: `uv run nemotron \u003Cfamily> --help`, `uv run nemotron \u003Cfamily> \u003Cstage> -c \u003Cprofile> -d`\n\n## Safe Workflow\n\n1. Gather only context relevant to the task: recipe family, selected profile, corpus path, existing SDG\u002Ftraining\u002Feval data, target stage range, artifact root, checkpoint path, execution mode, GPU IDs, and whether required secrets are configured. Never ask users to paste secret values.\n2. Start with cheap checks before expensive work:\n   - `uv run nemotron \u003Cfamily> --help`\n   - `uv run nemotron \u003Cfamily> \u003Cstage> --help`\n   - `uv run nemotron \u003Cfamily> \u003Cstage> -c \u003Cprofile> -d`\n   - `uv run nemotron \u003Cfamily> run -c \u003Cprofile> -d --from \u003Cstage> --to \u003Cstage>`\n   - `run --help` may omit inherited `-c` and `-d` options even though `run -c default -d ...` works; validate by running the dry-run when unsure.\n   - In an already prepared checkout, `uv run --no-sync ... --help` or `uv run --no-sync ... -d` can avoid unexpected dependency sync during read-only checks.\n3. Check prerequisites for the requested stage:\n   - Repo environment: `uv sync --all-extras` or the smallest relevant extra if documented by the repo.\n   - Stage 0 SDG: `NVIDIA_API_KEY`.\n   - Stages 1–3 GPU work: CUDA\u002FNVIDIA driver availability and enough VRAM.\n   - Stage 4 export: the NeMo Export-Deploy container when using TensorRT. Default Nemotron 3 Embed skips this stage.\n   - Stage 5 deploy: Docker plus the selected backend's image and artifact contract; default Nemotron 3 may use the checked-in vLLM image without NIM credentials. Load the family reference before requiring NGC credentials.\n   - Remote execution: root `env.toml` profile for `--run` or `--batch`; load `references\u002Fremote.md` when remote scheduling, logs, or GPU placement matter.\n4. Use dotlist overrides instead of editing defaults unless the user asks for reusable config changes. Keep the selected profile, artifact root, sequence length, prefixes, pooling\u002Fnormalization, prompt templates, and hard-negative counts consistent across stages.\n5. Avoid launching API, GPU, Docker, Slurm, NIM, or long-running jobs unless the user explicitly asked to run them. Offer or run dry-runs, config review, and small pilots first.\n6. For local execution, scope requested GPU IDs with `CUDA_VISIBLE_DEVICES=\u003Cids>`. For `--run` or `--batch`, configure scheduler resources such as `gpus_per_node` in the selected `env.toml` profile and let the scheduler assign devices; do not assume submit-shell `CUDA_VISIBLE_DEVICES` propagates remotely.\n7. For multi-stage local runs, prefer `uv run nemotron \u003Cfamily> run -c \u003Cprofile> --from \u003Cstage> --to \u003Cstage>`. Use `default` for rerank. The default `run` target stops at `eval`; `export` and `deploy` are opt-in.\n8. When evaluating quality, compare against the base model on a fixed held-out evaluation set before recommending deployment. Do not substitute a standalone public-benchmark eval for the recipe's own Stage 3 evaluation.\n9. For long-running SDG, prep, finetune, or eval work, start the process in a session-safe way and poll at human-scale intervals: roughly 60 seconds for small pilots and 120-300 seconds for larger runs.\n10. For failures, localize the failing stage, then inspect the stage config, expected inputs, output directory, and corresponding CLI wrapper or `run_uv.py`.\n\n## References\n\n- `references\u002Fembed.md`: embedding recipe stages, commands, defaults, output paths, and operating patterns.\n- `references\u002Frerank.md`: rerank recipe stages, commands, defaults, output paths, and operating patterns.\n- `references\u002Fevaluation.md`: metric interpretation, comparison hygiene, and deployment readiness checks.\n- `references\u002Fremote.md`: remote execution profiles, batch\u002Frun mode, GPU scoping, logs, and polling.\n\n## Examples\n\nUser asks: \"Recall is decent, but nDCG is poor and the right passage is around rank 40. Should I tune embed or rerank?\"\n\nLoad `references\u002Frerank.md` and `references\u002Fevaluation.md`, explain that acceptable recall with poor top-rank ordering points to reranker tuning, then offer a cheap preview before training.\n\n```bash\nuv run nemotron rerank run -c default -d --from prep --to eval\n```\n\n## Troubleshooting\n\nLocalize the failing stage, then inspect the stage config, expected inputs, output directory, and corresponding CLI wrapper or `run_uv.py`.\n\n## Limitations\n\n- Bundled references are condensed snapshots; verify commands, flags, defaults, and output paths against the active checkout before execution.\n- This skill does not provide datasets, checkpoints, credentials, GPU capacity, Docker images, or NIM services.\n\n## Output Style\n\nFor planning or debugging recommendations, use this shape when it helps: `Decision`, `Why`, `Required inputs`, `Preview command`, `Execution command`, `Avoid`, and `Next step`. Omit fields that are irrelevant to a short answer.\n\nGive concrete commands and file paths. State assumptions, expected inputs, expected outputs, and the cheapest validation step that proves the next action is ready. For long-running stages, separate preview commands from execution commands so the user can choose deliberately.\n\nWhen reporting a dry-run or real run, include a compact run report: command, mode, config, dotlist overrides, input paths, output paths, validation signal or metric file, and next cheapest check. Include the checkout commit when it is available.\n",{"data":35,"body":49},{"name":4,"version":36,"author":37,"license":26,"tags":38,"metadata":44,"tools":46,"description":6},"0.2.0","NVIDIA Nemotron Team \u003Cnoreply@nvidia.com>",[39,40,41,42,43],"nemotron","retrieval","fine-tuning","embeddings","reranking",{"author":37,"tags":45},[39,40,41,42,43],[47,48,21],"Read","Bash",{"type":50,"children":51},"root",[52,60,75,82,87,92,113,119,131,137,142,167,172,178,290,296,517,523,832,838,882,888,893,911,989,995,1006,1012,1025,1031,1087,1092,1097],{"type":53,"tag":54,"props":55,"children":56},"element","h1",{"id":4},[57],{"type":58,"value":59},"text","Nemotron Retrieval Recipes",{"type":53,"tag":61,"props":62,"children":63},"p",{},[64,66,73],{"type":58,"value":65},"Invocation: ",{"type":53,"tag":67,"props":68,"children":70},"code",{"className":69},[],[71],{"type":58,"value":72},"$nemotron-retrieval-recipes",{"type":58,"value":74},".",{"type":53,"tag":76,"props":77,"children":79},"h2",{"id":78},"purpose",[80],{"type":58,"value":81},"Purpose",{"type":53,"tag":61,"props":83,"children":84},{},[85],{"type":58,"value":86},"Use this skill to work with public Nemotron embedding and reranking retrieval recipes in a source checkout or installed package. Prefer the current checkout over memory, because the recipe CLI, configs, containers, and output paths are actively changing. Treat each recipe family as available only after its recipe directory and matching CLI files are present.",{"type":53,"tag":61,"props":88,"children":89},{},[90],{"type":58,"value":91},"This is a public product skill, not contributor-only guidance. Its value over static docs is to make an agent route the user's retrieval failure to the right recipe family, reconcile docs with the current checkout, avoid accidental long-running launches, preserve secrets, and return concrete preview\u002Fexecution\u002Frun-report commands.",{"type":53,"tag":61,"props":93,"children":94},{},[95,97,103,105,111],{"type":58,"value":96},"Use it only for tasks tied to the public Nemotron ",{"type":53,"tag":67,"props":98,"children":100},{"className":99},[],[101],{"type":58,"value":102},"embed",{"type":58,"value":104}," or ",{"type":53,"tag":67,"props":106,"children":108},{"className":107},[],[109],{"type":58,"value":110},"rerank",{"type":58,"value":112}," recipe flow. If the request is unrelated retrieval theory, generic vector database selection, generic benchmark advice, or non-recipe Docker\u002FSlurm\u002FNIM troubleshooting, stop with a short scope note and do not inspect recipe files in that turn.",{"type":53,"tag":76,"props":114,"children":116},{"id":115},"security-notes",[117],{"type":58,"value":118},"Security Notes",{"type":53,"tag":61,"props":120,"children":121},{},[122,124,129],{"type":58,"value":123},"Use ",{"type":53,"tag":67,"props":125,"children":127},{"className":126},[],[128],{"type":58,"value":48},{"type":58,"value":130}," for repo-scoped inspection, help, dry-run, and user-approved execution commands. Do not run API, GPU, Docker, Slurm, NIM, or other long-running work unless the user explicitly asks for it. Before Stage 0 SDG for either family, confirm the user's data-governance policy permits sending corpus content to the configured inference endpoints; otherwise use an approved private or air-gapped path. Never run broad environment dumps or commands that expose secret values. Prefer dotlist overrides and config review over editing recipe defaults.",{"type":53,"tag":76,"props":132,"children":134},{"id":133},"source-priority",[135],{"type":58,"value":136},"Source Priority",{"type":53,"tag":61,"props":138,"children":139},{},[140],{"type":58,"value":141},"Resolve conflicts in this order:",{"type":53,"tag":143,"props":144,"children":145},"ol",{},[146,152,157,162],{"type":53,"tag":147,"props":148,"children":149},"li",{},[150],{"type":58,"value":151},"Current checkout recipe, CLI, config, and source files.",{"type":53,"tag":147,"props":153,"children":154},{},[155],{"type":58,"value":156},"Bundled references in this skill.",{"type":53,"tag":147,"props":158,"children":159},{},[160],{"type":58,"value":161},"User-provided docs or saved snippets.",{"type":53,"tag":147,"props":163,"children":164},{},[165],{"type":58,"value":166},"Memory.",{"type":53,"tag":61,"props":168,"children":169},{},[170],{"type":58,"value":171},"For runnable commands, treat the current checkout as authoritative. If a required recipe directory, CLI command, config, or env profile is missing, report the blocker instead of guessing.",{"type":53,"tag":76,"props":173,"children":175},{"id":174},"prerequisites",[176],{"type":58,"value":177},"Prerequisites",{"type":53,"tag":179,"props":180,"children":181},"ul",{},[182,195,208,213,218,254],{"type":53,"tag":147,"props":183,"children":184},{},[185,187,193],{"type":58,"value":186},"Repo environment: ",{"type":53,"tag":67,"props":188,"children":190},{"className":189},[],[191],{"type":58,"value":192},"uv sync --all-extras",{"type":58,"value":194}," or the smallest relevant extra documented by the checkout.",{"type":53,"tag":147,"props":196,"children":197},{},[198,200,206],{"type":58,"value":199},"Stage 0 SDG: ",{"type":53,"tag":67,"props":201,"children":203},{"className":202},[],[204],{"type":58,"value":205},"NVIDIA_API_KEY",{"type":58,"value":207},"; never ask users to paste secret values.",{"type":53,"tag":147,"props":209,"children":210},{},[211],{"type":58,"value":212},"Stages 1–3 GPU work: CUDA\u002FNVIDIA driver availability and enough VRAM.",{"type":53,"tag":147,"props":214,"children":215},{},[216],{"type":58,"value":217},"Stage 4 export: NeMo Export-Deploy container when using TensorRT. The default Nemotron 3 Embed profile intentionally skips export.",{"type":53,"tag":147,"props":219,"children":220},{},[221,223,229,231,237,239,245,247,253],{"type":58,"value":222},"Stage 5 deploy: Docker. Default Nemotron 3 Embed can use the checked-in vLLM path with ",{"type":53,"tag":67,"props":224,"children":226},{"className":225},[],[227],{"type":58,"value":228},"backend=vllm",{"type":58,"value":230},", or a compatible ",{"type":53,"tag":67,"props":232,"children":234},{"className":233},[],[235],{"type":58,"value":236},"NEMOTRON3_EMBED_NIM_IMAGE",{"type":58,"value":238}," with ",{"type":53,"tag":67,"props":240,"children":242},{"className":241},[],[243],{"type":58,"value":244},"backend=nim",{"type":58,"value":246},"; Llama Embed and rerank deployment may require NGC access and ",{"type":53,"tag":67,"props":248,"children":250},{"className":249},[],[251],{"type":58,"value":252},"NGC_API_KEY",{"type":58,"value":74},{"type":53,"tag":147,"props":255,"children":256},{},[257,259,265,267,273,274,280,282,288],{"type":58,"value":258},"Remote execution: root ",{"type":53,"tag":67,"props":260,"children":262},{"className":261},[],[263],{"type":58,"value":264},"env.toml",{"type":58,"value":266}," profile for ",{"type":53,"tag":67,"props":268,"children":270},{"className":269},[],[271],{"type":58,"value":272},"--run",{"type":58,"value":104},{"type":53,"tag":67,"props":275,"children":277},{"className":276},[],[278],{"type":58,"value":279},"--batch",{"type":58,"value":281},"; load ",{"type":53,"tag":67,"props":283,"children":285},{"className":284},[],[286],{"type":58,"value":287},"references\u002Fremote.md",{"type":58,"value":289}," when remote scheduling, logs, or GPU placement matter.",{"type":53,"tag":76,"props":291,"children":293},{"id":292},"instructions",[294],{"type":58,"value":295},"Instructions",{"type":53,"tag":143,"props":297,"children":298},{},[299,350,429,452,457],{"type":53,"tag":147,"props":300,"children":301},{},[302,304],{"type":58,"value":303},"Identify the recipe family.\n",{"type":53,"tag":179,"props":305,"children":306},{},[307,326,345],{"type":53,"tag":147,"props":308,"children":309},{},[310,311,317,319,325],{"type":58,"value":123},{"type":53,"tag":67,"props":312,"children":314},{"className":313},[],[315],{"type":58,"value":316},"references\u002Fembed.md",{"type":58,"value":318}," for embedding, embed, bi-encoder, vector search, first-stage retrieval, low Recall@k, missing relevant documents, NIM embeddings, or ",{"type":53,"tag":67,"props":320,"children":322},{"className":321},[],[323],{"type":58,"value":324},"nemotron embed",{"type":58,"value":74},{"type":53,"tag":147,"props":327,"children":328},{},[329,330,336,338,344],{"type":58,"value":123},{"type":53,"tag":67,"props":331,"children":333},{"className":332},[],[334],{"type":58,"value":335},"references\u002Frerank.md",{"type":58,"value":337}," for rerank, reranker, cross-encoder, second-stage retrieval, acceptable recall but poor top-rank ordering, low nDCG with good Recall, or ",{"type":53,"tag":67,"props":339,"children":341},{"className":340},[],[342],{"type":58,"value":343},"nemotron rerank",{"type":58,"value":74},{"type":53,"tag":147,"props":346,"children":347},{},[348],{"type":58,"value":349},"Use both references only when the user asks about both families or asks which family to choose.",{"type":53,"tag":147,"props":351,"children":352},{},[353,355,360,362],{"type":58,"value":354},"For ",{"type":53,"tag":67,"props":356,"children":358},{"className":357},[],[359],{"type":58,"value":102},{"type":58,"value":361},", choose one model profile before composing stage commands.\n",{"type":53,"tag":179,"props":363,"children":364},{},[365,378,397,416],{"type":53,"tag":147,"props":366,"children":367},{},[368,370,376],{"type":58,"value":369},"Run ",{"type":53,"tag":67,"props":371,"children":373},{"className":372},[],[374],{"type":58,"value":375},"uv run nemotron embed info",{"type":58,"value":377}," when the requested model is unclear.",{"type":53,"tag":147,"props":379,"children":380},{},[381,382,388,390,396],{"type":58,"value":123},{"type":53,"tag":67,"props":383,"children":385},{"className":384},[],[386],{"type":58,"value":387},"-c default",{"type":58,"value":389}," for ",{"type":53,"tag":67,"props":391,"children":393},{"className":392},[],[394],{"type":58,"value":395},"nvidia\u002FNemotron-3-Embed-1B-BF16",{"type":58,"value":74},{"type":53,"tag":147,"props":398,"children":399},{},[400,401,407,408,414],{"type":58,"value":123},{"type":53,"tag":67,"props":402,"children":404},{"className":403},[],[405],{"type":58,"value":406},"-c llama",{"type":58,"value":389},{"type":53,"tag":67,"props":409,"children":411},{"className":410},[],[412],{"type":58,"value":413},"nvidia\u002Fllama-nemotron-embed-1b-v2",{"type":58,"value":415}," and its export path.",{"type":53,"tag":147,"props":417,"children":418},{},[419,421,427],{"type":58,"value":420},"Carry the selected profile and ",{"type":53,"tag":67,"props":422,"children":424},{"className":423},[],[425],{"type":58,"value":426},"artifact_root",{"type":58,"value":428}," through every stage; never combine artifacts from the two profiles.",{"type":53,"tag":147,"props":430,"children":431},{},[432,434],{"type":58,"value":433},"Choose the model family to tune from the retrieval failure mode.\n",{"type":53,"tag":179,"props":435,"children":436},{},[437,442,447],{"type":53,"tag":147,"props":438,"children":439},{},[440],{"type":58,"value":441},"Prefer embedding fine-tuning when relevant documents are absent from the candidate set.",{"type":53,"tag":147,"props":443,"children":444},{},[445],{"type":58,"value":446},"Prefer reranker fine-tuning when relevant documents are retrieved but ordered poorly near the top.",{"type":53,"tag":147,"props":448,"children":449},{},[450],{"type":58,"value":451},"For production retrieval stacks, remember that these are complementary: embed first, rerank candidates second.",{"type":53,"tag":147,"props":453,"children":454},{},[455],{"type":58,"value":456},"Identify the intent: plan a run, execute a stage, debug a failure, tune hyperparameters, interpret metrics, export\u002Fdeploy a model, inspect configs, or propose dotlist overrides.",{"type":53,"tag":147,"props":458,"children":459},{},[460,462],{"type":58,"value":461},"Inspect the current public surface before acting:\n",{"type":53,"tag":179,"props":463,"children":464},{},[465,476,487,498],{"type":53,"tag":147,"props":466,"children":467},{},[468,470],{"type":58,"value":469},"Recipe files: ",{"type":53,"tag":67,"props":471,"children":473},{"className":472},[],[474],{"type":58,"value":475},"src\u002Fnemotron\u002Frecipes\u002F\u003Cembed|rerank>\u002F",{"type":53,"tag":147,"props":477,"children":478},{},[479,481],{"type":58,"value":480},"CLI files: ",{"type":53,"tag":67,"props":482,"children":484},{"className":483},[],[485],{"type":58,"value":486},"src\u002Fnemotron\u002Fcli\u002Fcommands\u002F\u003Cembed|rerank>\u002F",{"type":53,"tag":147,"props":488,"children":489},{},[490,492],{"type":58,"value":491},"Configs: ",{"type":53,"tag":67,"props":493,"children":495},{"className":494},[],[496],{"type":58,"value":497},"src\u002Fnemotron\u002Frecipes\u002F\u003Cfamily>\u002Fstage*\u002Fconfig\u002F\u003Cprofile>.yaml",{"type":53,"tag":147,"props":499,"children":500},{},[501,503,509,511],{"type":58,"value":502},"Help and dry runs: ",{"type":53,"tag":67,"props":504,"children":506},{"className":505},[],[507],{"type":58,"value":508},"uv run nemotron \u003Cfamily> --help",{"type":58,"value":510},", ",{"type":53,"tag":67,"props":512,"children":514},{"className":513},[],[515],{"type":58,"value":516},"uv run nemotron \u003Cfamily> \u003Cstage> -c \u003Cprofile> -d",{"type":53,"tag":76,"props":518,"children":520},{"id":519},"safe-workflow",[521],{"type":58,"value":522},"Safe Workflow",{"type":53,"tag":143,"props":524,"children":525},{},[526,531,628,699,704,709,758,810,815,820],{"type":53,"tag":147,"props":527,"children":528},{},[529],{"type":58,"value":530},"Gather only context relevant to the task: recipe family, selected profile, corpus path, existing SDG\u002Ftraining\u002Feval data, target stage range, artifact root, checkpoint path, execution mode, GPU IDs, and whether required secrets are configured. Never ask users to paste secret values.",{"type":53,"tag":147,"props":532,"children":533},{},[534,536],{"type":58,"value":535},"Start with cheap checks before expensive work:\n",{"type":53,"tag":179,"props":537,"children":538},{},[539,547,556,564,573,608],{"type":53,"tag":147,"props":540,"children":541},{},[542],{"type":53,"tag":67,"props":543,"children":545},{"className":544},[],[546],{"type":58,"value":508},{"type":53,"tag":147,"props":548,"children":549},{},[550],{"type":53,"tag":67,"props":551,"children":553},{"className":552},[],[554],{"type":58,"value":555},"uv run nemotron \u003Cfamily> \u003Cstage> --help",{"type":53,"tag":147,"props":557,"children":558},{},[559],{"type":53,"tag":67,"props":560,"children":562},{"className":561},[],[563],{"type":58,"value":516},{"type":53,"tag":147,"props":565,"children":566},{},[567],{"type":53,"tag":67,"props":568,"children":570},{"className":569},[],[571],{"type":58,"value":572},"uv run nemotron \u003Cfamily> run -c \u003Cprofile> -d --from \u003Cstage> --to \u003Cstage>",{"type":53,"tag":147,"props":574,"children":575},{},[576,582,584,590,592,598,600,606],{"type":53,"tag":67,"props":577,"children":579},{"className":578},[],[580],{"type":58,"value":581},"run --help",{"type":58,"value":583}," may omit inherited ",{"type":53,"tag":67,"props":585,"children":587},{"className":586},[],[588],{"type":58,"value":589},"-c",{"type":58,"value":591}," and ",{"type":53,"tag":67,"props":593,"children":595},{"className":594},[],[596],{"type":58,"value":597},"-d",{"type":58,"value":599}," options even though ",{"type":53,"tag":67,"props":601,"children":603},{"className":602},[],[604],{"type":58,"value":605},"run -c default -d ...",{"type":58,"value":607}," works; validate by running the dry-run when unsure.",{"type":53,"tag":147,"props":609,"children":610},{},[611,613,619,620,626],{"type":58,"value":612},"In an already prepared checkout, ",{"type":53,"tag":67,"props":614,"children":616},{"className":615},[],[617],{"type":58,"value":618},"uv run --no-sync ... --help",{"type":58,"value":104},{"type":53,"tag":67,"props":621,"children":623},{"className":622},[],[624],{"type":58,"value":625},"uv run --no-sync ... -d",{"type":58,"value":627}," can avoid unexpected dependency sync during read-only checks.",{"type":53,"tag":147,"props":629,"children":630},{},[631,633],{"type":58,"value":632},"Check prerequisites for the requested stage:\n",{"type":53,"tag":179,"props":634,"children":635},{},[636,647,657,661,666,671],{"type":53,"tag":147,"props":637,"children":638},{},[639,640,645],{"type":58,"value":186},{"type":53,"tag":67,"props":641,"children":643},{"className":642},[],[644],{"type":58,"value":192},{"type":58,"value":646}," or the smallest relevant extra if documented by the repo.",{"type":53,"tag":147,"props":648,"children":649},{},[650,651,656],{"type":58,"value":199},{"type":53,"tag":67,"props":652,"children":654},{"className":653},[],[655],{"type":58,"value":205},{"type":58,"value":74},{"type":53,"tag":147,"props":658,"children":659},{},[660],{"type":58,"value":212},{"type":53,"tag":147,"props":662,"children":663},{},[664],{"type":58,"value":665},"Stage 4 export: the NeMo Export-Deploy container when using TensorRT. Default Nemotron 3 Embed skips this stage.",{"type":53,"tag":147,"props":667,"children":668},{},[669],{"type":58,"value":670},"Stage 5 deploy: Docker plus the selected backend's image and artifact contract; default Nemotron 3 may use the checked-in vLLM image without NIM credentials. Load the family reference before requiring NGC credentials.",{"type":53,"tag":147,"props":672,"children":673},{},[674,675,680,681,686,687,692,693,698],{"type":58,"value":258},{"type":53,"tag":67,"props":676,"children":678},{"className":677},[],[679],{"type":58,"value":264},{"type":58,"value":266},{"type":53,"tag":67,"props":682,"children":684},{"className":683},[],[685],{"type":58,"value":272},{"type":58,"value":104},{"type":53,"tag":67,"props":688,"children":690},{"className":689},[],[691],{"type":58,"value":279},{"type":58,"value":281},{"type":53,"tag":67,"props":694,"children":696},{"className":695},[],[697],{"type":58,"value":287},{"type":58,"value":289},{"type":53,"tag":147,"props":700,"children":701},{},[702],{"type":58,"value":703},"Use dotlist overrides instead of editing defaults unless the user asks for reusable config changes. Keep the selected profile, artifact root, sequence length, prefixes, pooling\u002Fnormalization, prompt templates, and hard-negative counts consistent across stages.",{"type":53,"tag":147,"props":705,"children":706},{},[707],{"type":58,"value":708},"Avoid launching API, GPU, Docker, Slurm, NIM, or long-running jobs unless the user explicitly asked to run them. Offer or run dry-runs, config review, and small pilots first.",{"type":53,"tag":147,"props":710,"children":711},{},[712,714,720,722,727,728,733,735,741,743,748,750,756],{"type":58,"value":713},"For local execution, scope requested GPU IDs with ",{"type":53,"tag":67,"props":715,"children":717},{"className":716},[],[718],{"type":58,"value":719},"CUDA_VISIBLE_DEVICES=\u003Cids>",{"type":58,"value":721},". For ",{"type":53,"tag":67,"props":723,"children":725},{"className":724},[],[726],{"type":58,"value":272},{"type":58,"value":104},{"type":53,"tag":67,"props":729,"children":731},{"className":730},[],[732],{"type":58,"value":279},{"type":58,"value":734},", configure scheduler resources such as ",{"type":53,"tag":67,"props":736,"children":738},{"className":737},[],[739],{"type":58,"value":740},"gpus_per_node",{"type":58,"value":742}," in the selected ",{"type":53,"tag":67,"props":744,"children":746},{"className":745},[],[747],{"type":58,"value":264},{"type":58,"value":749}," profile and let the scheduler assign devices; do not assume submit-shell ",{"type":53,"tag":67,"props":751,"children":753},{"className":752},[],[754],{"type":58,"value":755},"CUDA_VISIBLE_DEVICES",{"type":58,"value":757}," propagates remotely.",{"type":53,"tag":147,"props":759,"children":760},{},[761,763,769,771,777,779,785,787,793,795,801,802,808],{"type":58,"value":762},"For multi-stage local runs, prefer ",{"type":53,"tag":67,"props":764,"children":766},{"className":765},[],[767],{"type":58,"value":768},"uv run nemotron \u003Cfamily> run -c \u003Cprofile> --from \u003Cstage> --to \u003Cstage>",{"type":58,"value":770},". Use ",{"type":53,"tag":67,"props":772,"children":774},{"className":773},[],[775],{"type":58,"value":776},"default",{"type":58,"value":778}," for rerank. The default ",{"type":53,"tag":67,"props":780,"children":782},{"className":781},[],[783],{"type":58,"value":784},"run",{"type":58,"value":786}," target stops at ",{"type":53,"tag":67,"props":788,"children":790},{"className":789},[],[791],{"type":58,"value":792},"eval",{"type":58,"value":794},"; ",{"type":53,"tag":67,"props":796,"children":798},{"className":797},[],[799],{"type":58,"value":800},"export",{"type":58,"value":591},{"type":53,"tag":67,"props":803,"children":805},{"className":804},[],[806],{"type":58,"value":807},"deploy",{"type":58,"value":809}," are opt-in.",{"type":53,"tag":147,"props":811,"children":812},{},[813],{"type":58,"value":814},"When evaluating quality, compare against the base model on a fixed held-out evaluation set before recommending deployment. Do not substitute a standalone public-benchmark eval for the recipe's own Stage 3 evaluation.",{"type":53,"tag":147,"props":816,"children":817},{},[818],{"type":58,"value":819},"For long-running SDG, prep, finetune, or eval work, start the process in a session-safe way and poll at human-scale intervals: roughly 60 seconds for small pilots and 120-300 seconds for larger runs.",{"type":53,"tag":147,"props":821,"children":822},{},[823,825,831],{"type":58,"value":824},"For failures, localize the failing stage, then inspect the stage config, expected inputs, output directory, and corresponding CLI wrapper or ",{"type":53,"tag":67,"props":826,"children":828},{"className":827},[],[829],{"type":58,"value":830},"run_uv.py",{"type":58,"value":74},{"type":53,"tag":76,"props":833,"children":835},{"id":834},"references",[836],{"type":58,"value":837},"References",{"type":53,"tag":179,"props":839,"children":840},{},[841,851,861,872],{"type":53,"tag":147,"props":842,"children":843},{},[844,849],{"type":53,"tag":67,"props":845,"children":847},{"className":846},[],[848],{"type":58,"value":316},{"type":58,"value":850},": embedding recipe stages, commands, defaults, output paths, and operating patterns.",{"type":53,"tag":147,"props":852,"children":853},{},[854,859],{"type":53,"tag":67,"props":855,"children":857},{"className":856},[],[858],{"type":58,"value":335},{"type":58,"value":860},": rerank recipe stages, commands, defaults, output paths, and operating patterns.",{"type":53,"tag":147,"props":862,"children":863},{},[864,870],{"type":53,"tag":67,"props":865,"children":867},{"className":866},[],[868],{"type":58,"value":869},"references\u002Fevaluation.md",{"type":58,"value":871},": metric interpretation, comparison hygiene, and deployment readiness checks.",{"type":53,"tag":147,"props":873,"children":874},{},[875,880],{"type":53,"tag":67,"props":876,"children":878},{"className":877},[],[879],{"type":58,"value":287},{"type":58,"value":881},": remote execution profiles, batch\u002Frun mode, GPU scoping, logs, and polling.",{"type":53,"tag":76,"props":883,"children":885},{"id":884},"examples",[886],{"type":58,"value":887},"Examples",{"type":53,"tag":61,"props":889,"children":890},{},[891],{"type":58,"value":892},"User asks: \"Recall is decent, but nDCG is poor and the right passage is around rank 40. Should I tune embed or rerank?\"",{"type":53,"tag":61,"props":894,"children":895},{},[896,898,903,904,909],{"type":58,"value":897},"Load ",{"type":53,"tag":67,"props":899,"children":901},{"className":900},[],[902],{"type":58,"value":335},{"type":58,"value":591},{"type":53,"tag":67,"props":905,"children":907},{"className":906},[],[908],{"type":58,"value":869},{"type":58,"value":910},", explain that acceptable recall with poor top-rank ordering points to reranker tuning, then offer a cheap preview before training.",{"type":53,"tag":912,"props":913,"children":918},"pre",{"className":914,"code":915,"language":916,"meta":917,"style":917},"language-bash shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","uv run nemotron rerank run -c default -d --from prep --to eval\n","bash","",[919],{"type":53,"tag":67,"props":920,"children":921},{"__ignoreMap":917},[922],{"type":53,"tag":923,"props":924,"children":927},"span",{"class":925,"line":926},"line",1,[928,934,940,945,950,954,959,964,969,974,979,984],{"type":53,"tag":923,"props":929,"children":931},{"style":930},"--shiki-light:#E2931D;--shiki-default:#FFCB6B;--shiki-dark:#FFCB6B",[932],{"type":58,"value":933},"uv",{"type":53,"tag":923,"props":935,"children":937},{"style":936},"--shiki-light:#91B859;--shiki-default:#C3E88D;--shiki-dark:#C3E88D",[938],{"type":58,"value":939}," run",{"type":53,"tag":923,"props":941,"children":942},{"style":936},[943],{"type":58,"value":944}," nemotron",{"type":53,"tag":923,"props":946,"children":947},{"style":936},[948],{"type":58,"value":949}," rerank",{"type":53,"tag":923,"props":951,"children":952},{"style":936},[953],{"type":58,"value":939},{"type":53,"tag":923,"props":955,"children":956},{"style":936},[957],{"type":58,"value":958}," -c",{"type":53,"tag":923,"props":960,"children":961},{"style":936},[962],{"type":58,"value":963}," default",{"type":53,"tag":923,"props":965,"children":966},{"style":936},[967],{"type":58,"value":968}," -d",{"type":53,"tag":923,"props":970,"children":971},{"style":936},[972],{"type":58,"value":973}," --from",{"type":53,"tag":923,"props":975,"children":976},{"style":936},[977],{"type":58,"value":978}," prep",{"type":53,"tag":923,"props":980,"children":981},{"style":936},[982],{"type":58,"value":983}," --to",{"type":53,"tag":923,"props":985,"children":986},{"style":936},[987],{"type":58,"value":988}," eval\n",{"type":53,"tag":76,"props":990,"children":992},{"id":991},"troubleshooting",[993],{"type":58,"value":994},"Troubleshooting",{"type":53,"tag":61,"props":996,"children":997},{},[998,1000,1005],{"type":58,"value":999},"Localize the failing stage, then inspect the stage config, expected inputs, output directory, and corresponding CLI wrapper or ",{"type":53,"tag":67,"props":1001,"children":1003},{"className":1002},[],[1004],{"type":58,"value":830},{"type":58,"value":74},{"type":53,"tag":76,"props":1007,"children":1009},{"id":1008},"limitations",[1010],{"type":58,"value":1011},"Limitations",{"type":53,"tag":179,"props":1013,"children":1014},{},[1015,1020],{"type":53,"tag":147,"props":1016,"children":1017},{},[1018],{"type":58,"value":1019},"Bundled references are condensed snapshots; verify commands, flags, defaults, and output paths against the active checkout before execution.",{"type":53,"tag":147,"props":1021,"children":1022},{},[1023],{"type":58,"value":1024},"This skill does not provide datasets, checkpoints, credentials, GPU capacity, Docker images, or NIM services.",{"type":53,"tag":76,"props":1026,"children":1028},{"id":1027},"output-style",[1029],{"type":58,"value":1030},"Output Style",{"type":53,"tag":61,"props":1032,"children":1033},{},[1034,1036,1042,1043,1049,1050,1056,1057,1063,1064,1070,1071,1077,1079,1085],{"type":58,"value":1035},"For planning or debugging recommendations, use this shape when it helps: ",{"type":53,"tag":67,"props":1037,"children":1039},{"className":1038},[],[1040],{"type":58,"value":1041},"Decision",{"type":58,"value":510},{"type":53,"tag":67,"props":1044,"children":1046},{"className":1045},[],[1047],{"type":58,"value":1048},"Why",{"type":58,"value":510},{"type":53,"tag":67,"props":1051,"children":1053},{"className":1052},[],[1054],{"type":58,"value":1055},"Required inputs",{"type":58,"value":510},{"type":53,"tag":67,"props":1058,"children":1060},{"className":1059},[],[1061],{"type":58,"value":1062},"Preview command",{"type":58,"value":510},{"type":53,"tag":67,"props":1065,"children":1067},{"className":1066},[],[1068],{"type":58,"value":1069},"Execution command",{"type":58,"value":510},{"type":53,"tag":67,"props":1072,"children":1074},{"className":1073},[],[1075],{"type":58,"value":1076},"Avoid",{"type":58,"value":1078},", and ",{"type":53,"tag":67,"props":1080,"children":1082},{"className":1081},[],[1083],{"type":58,"value":1084},"Next step",{"type":58,"value":1086},". Omit fields that are irrelevant to a short answer.",{"type":53,"tag":61,"props":1088,"children":1089},{},[1090],{"type":58,"value":1091},"Give concrete commands and file paths. State assumptions, expected inputs, expected outputs, and the cheapest validation step that proves the next action is ready. For long-running stages, separate preview commands from execution commands so the user can choose deliberately.",{"type":53,"tag":61,"props":1093,"children":1094},{},[1095],{"type":58,"value":1096},"When reporting a dry-run or real run, include a compact run report: command, mode, config, dotlist overrides, input paths, output paths, validation signal or metric file, and next cheapest check. Include the checkout commit when it is available.",{"type":53,"tag":1098,"props":1099,"children":1100},"style",{},[1101],{"type":58,"value":1102},"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":1104,"total":1259},[1105,1121,1137,1148,1160,1174,1187,1201,1214,1225,1239,1248],{"slug":1106,"name":1106,"fn":1107,"description":1108,"org":1109,"tags":1110,"stars":1118,"repoUrl":1119,"updatedAt":1120},"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},[1111,1114,1117],{"name":1112,"slug":1113,"type":15},"Documentation","documentation",{"name":1115,"slug":1116,"type":15},"MCP","mcp",{"name":21,"slug":22,"type":15},21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":1122,"name":1122,"fn":1123,"description":1124,"org":1125,"tags":1126,"stars":1134,"repoUrl":1135,"updatedAt":1136},"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},[1127,1130,1131],{"name":1128,"slug":1129,"type":15},"Containers","containers",{"name":17,"slug":18,"type":15},{"name":1132,"slug":1133,"type":15},"Python","python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":1138,"name":1138,"fn":1139,"description":1140,"org":1141,"tags":1142,"stars":1134,"repoUrl":1135,"updatedAt":1147},"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},[1143,1146],{"name":1144,"slug":1145,"type":15},"CI\u002FCD","ci-cd",{"name":17,"slug":18,"type":15},"2026-07-14T05:25:59.97109",{"slug":1149,"name":1149,"fn":1150,"description":1151,"org":1152,"tags":1153,"stars":1134,"repoUrl":1135,"updatedAt":1159},"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},[1154,1155,1156],{"name":1144,"slug":1145,"type":15},{"name":17,"slug":18,"type":15},{"name":1157,"slug":1158,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":1161,"name":1161,"fn":1162,"description":1163,"org":1164,"tags":1165,"stars":1134,"repoUrl":1135,"updatedAt":1173},"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},[1166,1169,1170],{"name":1167,"slug":1168,"type":15},"Debugging","debugging",{"name":1157,"slug":1158,"type":15},{"name":1171,"slug":1172,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":1175,"name":1175,"fn":1176,"description":1177,"org":1178,"tags":1179,"stars":1134,"repoUrl":1135,"updatedAt":1186},"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},[1180,1183],{"name":1181,"slug":1182,"type":15},"Best Practices","best-practices",{"name":1184,"slug":1185,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":1188,"name":1188,"fn":1189,"description":1190,"org":1191,"tags":1192,"stars":1134,"repoUrl":1135,"updatedAt":1200},"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},[1193,1196,1199],{"name":1194,"slug":1195,"type":15},"Machine Learning","machine-learning",{"name":1197,"slug":1198,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":1202,"name":1202,"fn":1203,"description":1204,"org":1205,"tags":1206,"stars":1134,"repoUrl":1135,"updatedAt":1213},"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},[1207,1210],{"name":1208,"slug":1209,"type":15},"QA","qa",{"name":1211,"slug":1212,"type":15},"Testing","testing","2026-07-14T05:25:53.673039",{"slug":1215,"name":1215,"fn":1216,"description":1217,"org":1218,"tags":1219,"stars":1134,"repoUrl":1135,"updatedAt":1224},"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},[1220,1221],{"name":17,"slug":18,"type":15},{"name":1222,"slug":1223,"type":15},"Infrastructure","infrastructure","2026-07-14T05:25:49.362534",{"slug":1226,"name":1226,"fn":1227,"description":1228,"org":1229,"tags":1230,"stars":1134,"repoUrl":1135,"updatedAt":1238},"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},[1231,1234,1235],{"name":1232,"slug":1233,"type":15},"Code Review","code-review",{"name":1157,"slug":1158,"type":15},{"name":1236,"slug":1237,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":1240,"name":1240,"fn":1241,"description":1242,"org":1243,"tags":1244,"stars":1134,"repoUrl":1135,"updatedAt":1247},"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},[1245,1246],{"name":1208,"slug":1209,"type":15},{"name":1211,"slug":1212,"type":15},"2026-07-14T05:25:54.928983",{"slug":1249,"name":1249,"fn":1250,"description":1251,"org":1252,"tags":1253,"stars":1134,"repoUrl":1135,"updatedAt":1258},"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},[1254,1257],{"name":1255,"slug":1256,"type":15},"Automation","automation",{"name":1144,"slug":1145,"type":15},"2026-07-30T05:29:03.275638",496,{"items":1261,"total":1357},[1262,1279,1289,1303,1313,1328,1343],{"slug":1263,"name":1263,"fn":1264,"description":1265,"org":1266,"tags":1267,"stars":23,"repoUrl":24,"updatedAt":1278},"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},[1268,1271,1274,1275],{"name":1269,"slug":1270,"type":15},"Data Analysis","data-analysis",{"name":1272,"slug":1273,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":1276,"slug":1277,"type":15},"Performance","performance","2026-07-14T05:28:43.176466",{"slug":1280,"name":1280,"fn":1281,"description":1282,"org":1283,"tags":1284,"stars":23,"repoUrl":24,"updatedAt":1288},"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},[1285,1286,1287],{"name":17,"slug":18,"type":15},{"name":1222,"slug":1223,"type":15},{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":1290,"name":1290,"fn":1291,"description":1292,"org":1293,"tags":1294,"stars":23,"repoUrl":24,"updatedAt":1302},"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},[1295,1298,1299],{"name":1296,"slug":1297,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":1300,"slug":1301,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":1304,"name":1304,"fn":1305,"description":1306,"org":1307,"tags":1308,"stars":23,"repoUrl":24,"updatedAt":1312},"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},[1309,1310,1311],{"name":1269,"slug":1270,"type":15},{"name":9,"slug":8,"type":15},{"name":1211,"slug":1212,"type":15},"2026-07-17T05:29:03.913266",{"slug":1314,"name":1314,"fn":1315,"description":1316,"org":1317,"tags":1318,"stars":23,"repoUrl":24,"updatedAt":1327},"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},[1319,1320,1323,1324],{"name":1255,"slug":1256,"type":15},{"name":1321,"slug":1322,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":1325,"slug":1326,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":1329,"name":1329,"fn":1330,"description":1331,"org":1332,"tags":1333,"stars":23,"repoUrl":24,"updatedAt":1342},"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},[1334,1335,1338,1339],{"name":17,"slug":18,"type":15},{"name":1336,"slug":1337,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":1340,"slug":1341,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":1344,"name":1344,"fn":1345,"description":1346,"org":1347,"tags":1348,"stars":23,"repoUrl":24,"updatedAt":1356},"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},[1349,1350,1353],{"name":9,"slug":8,"type":15},{"name":1351,"slug":1352,"type":15},"Quantum Computing","quantum-computing",{"name":1354,"slug":1355,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305]