[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-huggingface-hf-cloud-sagemaker-deployment-planner":3,"mdc-cy06gg-key":35,"related-repo-huggingface-hf-cloud-sagemaker-deployment-planner":785,"related-org-huggingface-hf-cloud-sagemaker-deployment-planner":873},{"slug":4,"name":4,"fn":5,"description":6,"org":7,"tags":11,"stars":24,"repoUrl":25,"updatedAt":26,"license":27,"forks":28,"topics":29,"repo":30,"sourceUrl":33,"mdContent":34},"hf-cloud-sagemaker-deployment-planner","plan model deployments to Amazon SageMaker","Plan and coordinate the deployment of a model to Amazon SageMaker AI. Use this skill whenever the user wants to deploy, host, serve, or expose a model on SageMaker or AWS — including phrases like \"deploy a model\", \"host this LLM on AWS\", \"serve this embedding model\", \"deploy a reranker\", \"deploy a text-to-image \u002F diffusion model\", \"host this for async inference\", \"create an endpoint\", \"serve my fine-tuned model\", or any request that involves making a model available for inference on AWS. Use this even when the user is vague (e.g. \"I just want to get this running on AWS, you figure it out\"). Works for text-generation LLMs, embedding models, rerankers, classifiers, text-to-image \u002F diffusion models — picks the right serving stack and chooses between real-time and async inference. This is the entry-point skill for SageMaker deployment work — it asks clarifying questions, picks a deployment pathway, and coordinates the other deployment skills.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},"huggingface","Hugging Face","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fhuggingface.png",[12,15,18,21],{"name":9,"slug":13,"type":14},"hugging-face","tag",{"name":16,"slug":17,"type":14},"Deployment","deployment",{"name":19,"slug":20,"type":14},"AI Infrastructure","ai-infrastructure",{"name":22,"slug":23,"type":14},"AWS","aws",10861,"https:\u002F\u002Fgithub.com\u002Fhuggingface\u002Fskills","2026-07-08T05:55:37.387689",null,721,[],{"repoUrl":25,"stars":24,"forks":28,"topics":31,"description":32},[],"Give your agents the power of the Hugging Face ecosystem","https:\u002F\u002Fgithub.com\u002Fhuggingface\u002Fskills\u002Ftree\u002FHEAD\u002Fskills\u002Fhf-cloud-sagemaker-deployment-planner","---\nname: hf-cloud-sagemaker-deployment-planner\ndescription: 'Plan and coordinate the deployment of a model to Amazon SageMaker AI. Use this skill whenever the user wants to deploy, host, serve, or expose a model on SageMaker or AWS — including phrases like \"deploy a model\", \"host this LLM on AWS\", \"serve this embedding model\", \"deploy a reranker\", \"deploy a text-to-image \u002F diffusion model\", \"host this for async inference\", \"create an endpoint\", \"serve my fine-tuned model\", or any request that involves making a model available for inference on AWS. Use this even when the user is vague (e.g. \"I just want to get this running on AWS, you figure it out\"). Works for text-generation LLMs, embedding models, rerankers, classifiers, text-to-image \u002F diffusion models — picks the right serving stack and chooses between real-time and async inference. This is the entry-point skill for SageMaker deployment work — it asks clarifying questions, picks a deployment pathway, and coordinates the other deployment skills.'\n---\n\n# SageMaker Deployment Planner\n\nYou are helping a user deploy a model to Amazon SageMaker. Most users invoking this skill want the model deployed with reasonable defaults, in as few questions as possible. Ask only what you need, recommend a pathway honestly, and hand off to the specialized skills.\n\n## Workflow phases\n\n1. **Discovery** — what is being deployed and what are the constraints (this skill)\n2. **Pathway selection** — real-time \u002F serverless \u002F async \u002F batch \u002F Bedrock CMI (this skill)\n3. **Context preflight** — `hf-cloud-aws-context-discovery`, then `hf-cloud-python-env-setup`\n4. **IAM preflight** — `hf-cloud-sagemaker-iam-preflight`\n5. **Image selection** — `hf-cloud-serving-image-selection`\n6. **Deployment** — `hf-cloud-sagemaker-production-defaults`\n\nPhases 1–2 are this skill's job. The others activate when their patterns match.\n\n## Discovery: ask only what you need\n\nYou will eventually need to know:\n\n- **What model**: HuggingFace ID, S3 path to artifacts, or model name. If the user is vague (\"the model I fine-tuned\"), ask for the artifact location.\n- **Model type**: text-generation LLM, embedding\u002Freranker, or other (classifier, NER, etc.). This determines the serving stack — usually inferable from the model name (anything ending in `-embed-*`, starting with `BAAI\u002Fbge-`, `sentence-transformers\u002F*` etc. is embeddings; chat\u002Finstruct models are LLMs). Only ask if it's genuinely ambiguous.\n- **Traffic shape**: roughly how often will this be called?\n- **Latency tolerance**: interactive, near-real-time, or async?\n- **Cost sensitivity**: ask only if the user signals it or the traffic pattern is ambiguous.\n\nRegion comes from `hf-cloud-aws-context-discovery` — don't ask unless the user volunteers it.\n\nDo **not** front-load all of these. A common minimal set is just: *what model, and roughly how often will it be called?* The model name usually settles the model-type question. That alone is often enough to narrow the pathway to two candidates. If the user already told you something, don't ask again.\n\n## Pathway selection\n\n| Pathway | When it fits | When it does not |\n|---|---|---|\n| **Real-time endpoint** | Steady traffic, sub-second to few-second latency, always-on | Very spiky or very sparse traffic (wastes money on idle) |\n| **Serverless inference** | Spiky\u002Fintermittent, tolerates cold starts (~10s+), simpler models | LLMs above a few B params (memory\u002Fcold-start limits), strict SLAs |\n| **Async inference** | Long inference (>60s), large payloads, queue-friendly | Interactive synchronous calls |\n| **Batch transform** | Offline scoring over a dataset | Anything online or interactive |\n| **Bedrock Custom Model Import** | Wants Bedrock-compatible API, supported base family, weights only | Custom inference logic, unsupported architectures |\n\nFor LLMs, **real-time endpoints are the default** unless traffic is explicitly spiky\u002Fsparse or inference is long-running. Serverless looks attractive for \"low traffic\" cases but most LLMs exceed its memory limits.\n\nFor **embeddings**, real-time is again the default — but CPU instances are usually the right choice (much cheaper, fast enough for most embedding workloads). Don't reflexively recommend GPU instances for embedding models; ask `hf-cloud-serving-image-selection` to consider CPU variants if the model is small (\u003C1B params) and traffic is moderate.\n\nFor **text-to-image, video generation, or other long-inference workloads** (>30s per request) where traffic is also bursty: async inference is the right answer. It supports genuine scale-to-zero between batches and queues requests via S3, so you don't pay for idle GPU. `hf-cloud-sagemaker-production-defaults` has a dedicated `deploy_async.py` for this.\n\nReal-time and async are the two scripted pathways. Serverless, batch transform, and Bedrock Custom Model Import are not currently scripted — for those, hand the user off with a brief explanation rather than trying to deploy them through this workflow.\n\nIf two pathways are both reasonable, say so in one sentence each and pick one. Don't bury the recommendation in options.\n\n## Instance selection: check quota before recommending\n\nEndpoint quotas are per instance type, per region, and default to **0** for GPU types in many accounts. Recommending an instance the account can't launch wastes a full deploy cycle on `ResourceLimitExceeded`. Check first:\n\n```bash\naws service-quotas list-service-quotas --service-code sagemaker --region \u003Cregion> \\\n    --query \"Quotas[?contains(QuotaName, 'for endpoint usage') && Value > \\`0\\`].[QuotaName, Value]\" \\\n    --output table\n```\n\nIf the type you want isn't in the result, recommend one that is — or tell the user to request an increase (hours to days) *before* creating anything.\n\nGPU family notes for the common 24 GB tier:\n\n- `ml.g5.*` (A10G) and `ml.g6.*` (L4) both work with current vLLM images when the gpu-3-1 AMI is set (see `hf-cloud-serving-image-selection`). g6 is the newer generation and slightly cheaper per hour; g5 has roughly double the memory bandwidth, which usually means better LLM token throughput. Pick whichever has quota; when both do, either is defensible — g5 for throughput, g6 for cost.\n- `ml.g6e.*` (L40S, 48 GB) when the model doesn't fit in 24 GB.\n\nOnce you have enough to recommend, state it plainly:\n\n> Based on what you've told me, I'd recommend a real-time endpoint on `ml.g5.xlarge`. The model is small enough that this is cost-effective, and your traffic pattern is steady enough that you won't be paying for idle. Alternative: serverless would be cheaper if traffic dries up for hours at a time, but Qwen3-0.6B is at the edge of serverless memory limits and cold starts would be 15–30s. Want me to proceed with the real-time endpoint?\n\nThen wait for confirmation. The user should know what they're about to spend money on before you create anything.\n\nThe plan lives in the conversation — don't generate `plan.yaml` or similar artifacts unless explicitly asked.\n\n## Style\n\n- Users invoking this skill are deferring to the agent because they don't want to do AWS plumbing. Match that energy: efficient, not exhaustive.\n- One round of clarifying questions is usually enough. Three rounds is interrogation.\n- When you don't know something specific (current image URI, SDK API surface, quotas), check it rather than guess. Other skills handle the \"how to check\" details.\n- If the user pushes back on a recommendation, accept it. They know their constraints better than you do.\n",{"data":36,"body":37},{"name":4,"description":6},{"type":38,"children":39},"root",[40,49,55,62,157,162,168,173,251,263,283,288,426,438,457,483,488,493,499,519,654,666,671,711,716,733,738,751,757,780],{"type":41,"tag":42,"props":43,"children":45},"element","h1",{"id":44},"sagemaker-deployment-planner",[46],{"type":47,"value":48},"text","SageMaker Deployment Planner",{"type":41,"tag":50,"props":51,"children":52},"p",{},[53],{"type":47,"value":54},"You are helping a user deploy a model to Amazon SageMaker. Most users invoking this skill want the model deployed with reasonable defaults, in as few questions as possible. Ask only what you need, recommend a pathway honestly, and hand off to the specialized skills.",{"type":41,"tag":56,"props":57,"children":59},"h2",{"id":58},"workflow-phases",[60],{"type":47,"value":61},"Workflow phases",{"type":41,"tag":63,"props":64,"children":65},"ol",{},[66,78,88,113,128,143],{"type":41,"tag":67,"props":68,"children":69},"li",{},[70,76],{"type":41,"tag":71,"props":72,"children":73},"strong",{},[74],{"type":47,"value":75},"Discovery",{"type":47,"value":77}," — what is being deployed and what are the constraints (this skill)",{"type":41,"tag":67,"props":79,"children":80},{},[81,86],{"type":41,"tag":71,"props":82,"children":83},{},[84],{"type":47,"value":85},"Pathway selection",{"type":47,"value":87}," — real-time \u002F serverless \u002F async \u002F batch \u002F Bedrock CMI (this skill)",{"type":41,"tag":67,"props":89,"children":90},{},[91,96,98,105,107],{"type":41,"tag":71,"props":92,"children":93},{},[94],{"type":47,"value":95},"Context preflight",{"type":47,"value":97}," — ",{"type":41,"tag":99,"props":100,"children":102},"code",{"className":101},[],[103],{"type":47,"value":104},"hf-cloud-aws-context-discovery",{"type":47,"value":106},", then ",{"type":41,"tag":99,"props":108,"children":110},{"className":109},[],[111],{"type":47,"value":112},"hf-cloud-python-env-setup",{"type":41,"tag":67,"props":114,"children":115},{},[116,121,122],{"type":41,"tag":71,"props":117,"children":118},{},[119],{"type":47,"value":120},"IAM preflight",{"type":47,"value":97},{"type":41,"tag":99,"props":123,"children":125},{"className":124},[],[126],{"type":47,"value":127},"hf-cloud-sagemaker-iam-preflight",{"type":41,"tag":67,"props":129,"children":130},{},[131,136,137],{"type":41,"tag":71,"props":132,"children":133},{},[134],{"type":47,"value":135},"Image selection",{"type":47,"value":97},{"type":41,"tag":99,"props":138,"children":140},{"className":139},[],[141],{"type":47,"value":142},"hf-cloud-serving-image-selection",{"type":41,"tag":67,"props":144,"children":145},{},[146,150,151],{"type":41,"tag":71,"props":147,"children":148},{},[149],{"type":47,"value":16},{"type":47,"value":97},{"type":41,"tag":99,"props":152,"children":154},{"className":153},[],[155],{"type":47,"value":156},"hf-cloud-sagemaker-production-defaults",{"type":41,"tag":50,"props":158,"children":159},{},[160],{"type":47,"value":161},"Phases 1–2 are this skill's job. The others activate when their patterns match.",{"type":41,"tag":56,"props":163,"children":165},{"id":164},"discovery-ask-only-what-you-need",[166],{"type":47,"value":167},"Discovery: ask only what you need",{"type":41,"tag":50,"props":169,"children":170},{},[171],{"type":47,"value":172},"You will eventually need to know:",{"type":41,"tag":174,"props":175,"children":176},"ul",{},[177,187,221,231,241],{"type":41,"tag":67,"props":178,"children":179},{},[180,185],{"type":41,"tag":71,"props":181,"children":182},{},[183],{"type":47,"value":184},"What model",{"type":47,"value":186},": HuggingFace ID, S3 path to artifacts, or model name. If the user is vague (\"the model I fine-tuned\"), ask for the artifact location.",{"type":41,"tag":67,"props":188,"children":189},{},[190,195,197,203,205,211,213,219],{"type":41,"tag":71,"props":191,"children":192},{},[193],{"type":47,"value":194},"Model type",{"type":47,"value":196},": text-generation LLM, embedding\u002Freranker, or other (classifier, NER, etc.). This determines the serving stack — usually inferable from the model name (anything ending in ",{"type":41,"tag":99,"props":198,"children":200},{"className":199},[],[201],{"type":47,"value":202},"-embed-*",{"type":47,"value":204},", starting with ",{"type":41,"tag":99,"props":206,"children":208},{"className":207},[],[209],{"type":47,"value":210},"BAAI\u002Fbge-",{"type":47,"value":212},", ",{"type":41,"tag":99,"props":214,"children":216},{"className":215},[],[217],{"type":47,"value":218},"sentence-transformers\u002F*",{"type":47,"value":220}," etc. is embeddings; chat\u002Finstruct models are LLMs). Only ask if it's genuinely ambiguous.",{"type":41,"tag":67,"props":222,"children":223},{},[224,229],{"type":41,"tag":71,"props":225,"children":226},{},[227],{"type":47,"value":228},"Traffic shape",{"type":47,"value":230},": roughly how often will this be called?",{"type":41,"tag":67,"props":232,"children":233},{},[234,239],{"type":41,"tag":71,"props":235,"children":236},{},[237],{"type":47,"value":238},"Latency tolerance",{"type":47,"value":240},": interactive, near-real-time, or async?",{"type":41,"tag":67,"props":242,"children":243},{},[244,249],{"type":41,"tag":71,"props":245,"children":246},{},[247],{"type":47,"value":248},"Cost sensitivity",{"type":47,"value":250},": ask only if the user signals it or the traffic pattern is ambiguous.",{"type":41,"tag":50,"props":252,"children":253},{},[254,256,261],{"type":47,"value":255},"Region comes from ",{"type":41,"tag":99,"props":257,"children":259},{"className":258},[],[260],{"type":47,"value":104},{"type":47,"value":262}," — don't ask unless the user volunteers it.",{"type":41,"tag":50,"props":264,"children":265},{},[266,268,273,275,281],{"type":47,"value":267},"Do ",{"type":41,"tag":71,"props":269,"children":270},{},[271],{"type":47,"value":272},"not",{"type":47,"value":274}," front-load all of these. A common minimal set is just: ",{"type":41,"tag":276,"props":277,"children":278},"em",{},[279],{"type":47,"value":280},"what model, and roughly how often will it be called?",{"type":47,"value":282}," The model name usually settles the model-type question. That alone is often enough to narrow the pathway to two candidates. If the user already told you something, don't ask again.",{"type":41,"tag":56,"props":284,"children":286},{"id":285},"pathway-selection",[287],{"type":47,"value":85},{"type":41,"tag":289,"props":290,"children":291},"table",{},[292,316],{"type":41,"tag":293,"props":294,"children":295},"thead",{},[296],{"type":41,"tag":297,"props":298,"children":299},"tr",{},[300,306,311],{"type":41,"tag":301,"props":302,"children":303},"th",{},[304],{"type":47,"value":305},"Pathway",{"type":41,"tag":301,"props":307,"children":308},{},[309],{"type":47,"value":310},"When it fits",{"type":41,"tag":301,"props":312,"children":313},{},[314],{"type":47,"value":315},"When it does not",{"type":41,"tag":317,"props":318,"children":319},"tbody",{},[320,342,363,384,405],{"type":41,"tag":297,"props":321,"children":322},{},[323,332,337],{"type":41,"tag":324,"props":325,"children":326},"td",{},[327],{"type":41,"tag":71,"props":328,"children":329},{},[330],{"type":47,"value":331},"Real-time endpoint",{"type":41,"tag":324,"props":333,"children":334},{},[335],{"type":47,"value":336},"Steady traffic, sub-second to few-second latency, always-on",{"type":41,"tag":324,"props":338,"children":339},{},[340],{"type":47,"value":341},"Very spiky or very sparse traffic (wastes money on idle)",{"type":41,"tag":297,"props":343,"children":344},{},[345,353,358],{"type":41,"tag":324,"props":346,"children":347},{},[348],{"type":41,"tag":71,"props":349,"children":350},{},[351],{"type":47,"value":352},"Serverless inference",{"type":41,"tag":324,"props":354,"children":355},{},[356],{"type":47,"value":357},"Spiky\u002Fintermittent, tolerates cold starts (~10s+), simpler models",{"type":41,"tag":324,"props":359,"children":360},{},[361],{"type":47,"value":362},"LLMs above a few B params (memory\u002Fcold-start limits), strict SLAs",{"type":41,"tag":297,"props":364,"children":365},{},[366,374,379],{"type":41,"tag":324,"props":367,"children":368},{},[369],{"type":41,"tag":71,"props":370,"children":371},{},[372],{"type":47,"value":373},"Async inference",{"type":41,"tag":324,"props":375,"children":376},{},[377],{"type":47,"value":378},"Long inference (>60s), large payloads, queue-friendly",{"type":41,"tag":324,"props":380,"children":381},{},[382],{"type":47,"value":383},"Interactive synchronous calls",{"type":41,"tag":297,"props":385,"children":386},{},[387,395,400],{"type":41,"tag":324,"props":388,"children":389},{},[390],{"type":41,"tag":71,"props":391,"children":392},{},[393],{"type":47,"value":394},"Batch transform",{"type":41,"tag":324,"props":396,"children":397},{},[398],{"type":47,"value":399},"Offline scoring over a dataset",{"type":41,"tag":324,"props":401,"children":402},{},[403],{"type":47,"value":404},"Anything online or interactive",{"type":41,"tag":297,"props":406,"children":407},{},[408,416,421],{"type":41,"tag":324,"props":409,"children":410},{},[411],{"type":41,"tag":71,"props":412,"children":413},{},[414],{"type":47,"value":415},"Bedrock Custom Model Import",{"type":41,"tag":324,"props":417,"children":418},{},[419],{"type":47,"value":420},"Wants Bedrock-compatible API, supported base family, weights only",{"type":41,"tag":324,"props":422,"children":423},{},[424],{"type":47,"value":425},"Custom inference logic, unsupported architectures",{"type":41,"tag":50,"props":427,"children":428},{},[429,431,436],{"type":47,"value":430},"For LLMs, ",{"type":41,"tag":71,"props":432,"children":433},{},[434],{"type":47,"value":435},"real-time endpoints are the default",{"type":47,"value":437}," unless traffic is explicitly spiky\u002Fsparse or inference is long-running. Serverless looks attractive for \"low traffic\" cases but most LLMs exceed its memory limits.",{"type":41,"tag":50,"props":439,"children":440},{},[441,443,448,450,455],{"type":47,"value":442},"For ",{"type":41,"tag":71,"props":444,"children":445},{},[446],{"type":47,"value":447},"embeddings",{"type":47,"value":449},", real-time is again the default — but CPU instances are usually the right choice (much cheaper, fast enough for most embedding workloads). Don't reflexively recommend GPU instances for embedding models; ask ",{"type":41,"tag":99,"props":451,"children":453},{"className":452},[],[454],{"type":47,"value":142},{"type":47,"value":456}," to consider CPU variants if the model is small (\u003C1B params) and traffic is moderate.",{"type":41,"tag":50,"props":458,"children":459},{},[460,461,466,468,473,475,481],{"type":47,"value":442},{"type":41,"tag":71,"props":462,"children":463},{},[464],{"type":47,"value":465},"text-to-image, video generation, or other long-inference workloads",{"type":47,"value":467}," (>30s per request) where traffic is also bursty: async inference is the right answer. It supports genuine scale-to-zero between batches and queues requests via S3, so you don't pay for idle GPU. ",{"type":41,"tag":99,"props":469,"children":471},{"className":470},[],[472],{"type":47,"value":156},{"type":47,"value":474}," has a dedicated ",{"type":41,"tag":99,"props":476,"children":478},{"className":477},[],[479],{"type":47,"value":480},"deploy_async.py",{"type":47,"value":482}," for this.",{"type":41,"tag":50,"props":484,"children":485},{},[486],{"type":47,"value":487},"Real-time and async are the two scripted pathways. Serverless, batch transform, and Bedrock Custom Model Import are not currently scripted — for those, hand the user off with a brief explanation rather than trying to deploy them through this workflow.",{"type":41,"tag":50,"props":489,"children":490},{},[491],{"type":47,"value":492},"If two pathways are both reasonable, say so in one sentence each and pick one. Don't bury the recommendation in options.",{"type":41,"tag":56,"props":494,"children":496},{"id":495},"instance-selection-check-quota-before-recommending",[497],{"type":47,"value":498},"Instance selection: check quota before recommending",{"type":41,"tag":50,"props":500,"children":501},{},[502,504,509,511,517],{"type":47,"value":503},"Endpoint quotas are per instance type, per region, and default to ",{"type":41,"tag":71,"props":505,"children":506},{},[507],{"type":47,"value":508},"0",{"type":47,"value":510}," for GPU types in many accounts. Recommending an instance the account can't launch wastes a full deploy cycle on ",{"type":41,"tag":99,"props":512,"children":514},{"className":513},[],[515],{"type":47,"value":516},"ResourceLimitExceeded",{"type":47,"value":518},". Check first:",{"type":41,"tag":520,"props":521,"children":526},"pre",{"className":522,"code":523,"language":524,"meta":525,"style":525},"language-bash shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","aws service-quotas list-service-quotas --service-code sagemaker --region \u003Cregion> \\\n    --query \"Quotas[?contains(QuotaName, 'for endpoint usage') && Value > \\`0\\`].[QuotaName, Value]\" \\\n    --output table\n","bash","",[527],{"type":41,"tag":99,"props":528,"children":529},{"__ignoreMap":525},[530,594,640],{"type":41,"tag":531,"props":532,"children":535},"span",{"class":533,"line":534},"line",1,[536,541,547,552,557,562,567,573,578,584,589],{"type":41,"tag":531,"props":537,"children":539},{"style":538},"--shiki-light:#E2931D;--shiki-default:#FFCB6B;--shiki-dark:#FFCB6B",[540],{"type":47,"value":23},{"type":41,"tag":531,"props":542,"children":544},{"style":543},"--shiki-light:#91B859;--shiki-default:#C3E88D;--shiki-dark:#C3E88D",[545],{"type":47,"value":546}," service-quotas",{"type":41,"tag":531,"props":548,"children":549},{"style":543},[550],{"type":47,"value":551}," list-service-quotas",{"type":41,"tag":531,"props":553,"children":554},{"style":543},[555],{"type":47,"value":556}," --service-code",{"type":41,"tag":531,"props":558,"children":559},{"style":543},[560],{"type":47,"value":561}," sagemaker",{"type":41,"tag":531,"props":563,"children":564},{"style":543},[565],{"type":47,"value":566}," --region",{"type":41,"tag":531,"props":568,"children":570},{"style":569},"--shiki-light:#39ADB5;--shiki-default:#89DDFF;--shiki-dark:#89DDFF",[571],{"type":47,"value":572}," \u003C",{"type":41,"tag":531,"props":574,"children":575},{"style":543},[576],{"type":47,"value":577},"regio",{"type":41,"tag":531,"props":579,"children":581},{"style":580},"--shiki-light:#90A4AE;--shiki-default:#EEFFFF;--shiki-dark:#BABED8",[582],{"type":47,"value":583},"n",{"type":41,"tag":531,"props":585,"children":586},{"style":569},[587],{"type":47,"value":588},">",{"type":41,"tag":531,"props":590,"children":591},{"style":580},[592],{"type":47,"value":593}," \\\n",{"type":41,"tag":531,"props":595,"children":597},{"class":533,"line":596},2,[598,603,608,613,618,622,626,631,636],{"type":41,"tag":531,"props":599,"children":600},{"style":543},[601],{"type":47,"value":602},"    --query",{"type":41,"tag":531,"props":604,"children":605},{"style":569},[606],{"type":47,"value":607}," \"",{"type":41,"tag":531,"props":609,"children":610},{"style":543},[611],{"type":47,"value":612},"Quotas[?contains(QuotaName, 'for endpoint usage') && Value > ",{"type":41,"tag":531,"props":614,"children":615},{"style":580},[616],{"type":47,"value":617},"\\`",{"type":41,"tag":531,"props":619,"children":620},{"style":543},[621],{"type":47,"value":508},{"type":41,"tag":531,"props":623,"children":624},{"style":580},[625],{"type":47,"value":617},{"type":41,"tag":531,"props":627,"children":628},{"style":543},[629],{"type":47,"value":630},"].[QuotaName, Value]",{"type":41,"tag":531,"props":632,"children":633},{"style":569},[634],{"type":47,"value":635},"\"",{"type":41,"tag":531,"props":637,"children":638},{"style":580},[639],{"type":47,"value":593},{"type":41,"tag":531,"props":641,"children":643},{"class":533,"line":642},3,[644,649],{"type":41,"tag":531,"props":645,"children":646},{"style":543},[647],{"type":47,"value":648},"    --output",{"type":41,"tag":531,"props":650,"children":651},{"style":543},[652],{"type":47,"value":653}," table\n",{"type":41,"tag":50,"props":655,"children":656},{},[657,659,664],{"type":47,"value":658},"If the type you want isn't in the result, recommend one that is — or tell the user to request an increase (hours to days) ",{"type":41,"tag":276,"props":660,"children":661},{},[662],{"type":47,"value":663},"before",{"type":47,"value":665}," creating anything.",{"type":41,"tag":50,"props":667,"children":668},{},[669],{"type":47,"value":670},"GPU family notes for the common 24 GB tier:",{"type":41,"tag":174,"props":672,"children":673},{},[674,700],{"type":41,"tag":67,"props":675,"children":676},{},[677,683,685,691,693,698],{"type":41,"tag":99,"props":678,"children":680},{"className":679},[],[681],{"type":47,"value":682},"ml.g5.*",{"type":47,"value":684}," (A10G) and ",{"type":41,"tag":99,"props":686,"children":688},{"className":687},[],[689],{"type":47,"value":690},"ml.g6.*",{"type":47,"value":692}," (L4) both work with current vLLM images when the gpu-3-1 AMI is set (see ",{"type":41,"tag":99,"props":694,"children":696},{"className":695},[],[697],{"type":47,"value":142},{"type":47,"value":699},"). g6 is the newer generation and slightly cheaper per hour; g5 has roughly double the memory bandwidth, which usually means better LLM token throughput. Pick whichever has quota; when both do, either is defensible — g5 for throughput, g6 for cost.",{"type":41,"tag":67,"props":701,"children":702},{},[703,709],{"type":41,"tag":99,"props":704,"children":706},{"className":705},[],[707],{"type":47,"value":708},"ml.g6e.*",{"type":47,"value":710}," (L40S, 48 GB) when the model doesn't fit in 24 GB.",{"type":41,"tag":50,"props":712,"children":713},{},[714],{"type":47,"value":715},"Once you have enough to recommend, state it plainly:",{"type":41,"tag":717,"props":718,"children":719},"blockquote",{},[720],{"type":41,"tag":50,"props":721,"children":722},{},[723,725,731],{"type":47,"value":724},"Based on what you've told me, I'd recommend a real-time endpoint on ",{"type":41,"tag":99,"props":726,"children":728},{"className":727},[],[729],{"type":47,"value":730},"ml.g5.xlarge",{"type":47,"value":732},". The model is small enough that this is cost-effective, and your traffic pattern is steady enough that you won't be paying for idle. Alternative: serverless would be cheaper if traffic dries up for hours at a time, but Qwen3-0.6B is at the edge of serverless memory limits and cold starts would be 15–30s. Want me to proceed with the real-time endpoint?",{"type":41,"tag":50,"props":734,"children":735},{},[736],{"type":47,"value":737},"Then wait for confirmation. The user should know what they're about to spend money on before you create anything.",{"type":41,"tag":50,"props":739,"children":740},{},[741,743,749],{"type":47,"value":742},"The plan lives in the conversation — don't generate ",{"type":41,"tag":99,"props":744,"children":746},{"className":745},[],[747],{"type":47,"value":748},"plan.yaml",{"type":47,"value":750}," or similar artifacts unless explicitly asked.",{"type":41,"tag":56,"props":752,"children":754},{"id":753},"style",[755],{"type":47,"value":756},"Style",{"type":41,"tag":174,"props":758,"children":759},{},[760,765,770,775],{"type":41,"tag":67,"props":761,"children":762},{},[763],{"type":47,"value":764},"Users invoking this skill are deferring to the agent because they don't want to do AWS plumbing. Match that energy: efficient, not exhaustive.",{"type":41,"tag":67,"props":766,"children":767},{},[768],{"type":47,"value":769},"One round of clarifying questions is usually enough. Three rounds is interrogation.",{"type":41,"tag":67,"props":771,"children":772},{},[773],{"type":47,"value":774},"When you don't know something specific (current image URI, SDK API surface, quotas), check it rather than guess. Other skills handle the \"how to check\" details.",{"type":41,"tag":67,"props":776,"children":777},{},[778],{"type":47,"value":779},"If the user pushes back on a recommendation, accept it. They know their constraints better than you do.",{"type":41,"tag":753,"props":781,"children":782},{},[783],{"type":47,"value":784},"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":786,"total":872},[787,804,815,829,836,850,863],{"slug":788,"name":788,"fn":789,"description":790,"org":791,"tags":792,"stars":24,"repoUrl":25,"updatedAt":803},"hf-cli","manage Hugging Face Hub resources via CLI","Hugging Face Hub CLI (`hf`) for downloading, uploading, and managing models, datasets, spaces, buckets, repos, papers, jobs, and more on the Hugging Face Hub. Use when: handling authentication; managing local cache; managing Hugging Face Buckets; running or scheduling jobs on Hugging Face infrastructure; managing Hugging Face repos; discussions and pull requests; browsing models, datasets and spaces; reading, searching, or browsing academic papers; managing collections; querying datasets; configuring spaces; setting up webhooks; or deploying and managing HF Inference Endpoints. Make sure to use this skill whenever the user mentions 'hf', 'huggingface', 'Hugging Face', 'huggingface-cli', or 'hugging face cli', or wants to do anything related to the Hugging Face ecosystem and to AI and ML in general. Also use for cloud storage needs like training checkpoints, data pipelines, or agent traces. Use even if the user doesn't explicitly ask for a CLI command. Replaces the deprecated `huggingface-cli`.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[793,796,799,800],{"name":794,"slug":795,"type":14},"CLI","cli",{"name":797,"slug":798,"type":14},"Datasets","datasets",{"name":9,"slug":13,"type":14},{"name":801,"slug":802,"type":14},"LLM","llm","2026-04-06T18:25:34.020855",{"slug":104,"name":104,"fn":805,"description":806,"org":807,"tags":808,"stars":24,"repoUrl":25,"updatedAt":814},"discover local AWS environment context","Discover the user's local AWS context (active profile, region, account ID, caller identity) at the start of any AWS task. Use this skill before any other AWS work — deploying to SageMaker, creating resources, calling AWS APIs, or anything that touches an AWS account. Use it especially when the user has not specified a region or profile explicitly, when they say things like \"use my AWS account\", \"deploy to AWS\", \"use my profile\", or when about to make any AWS CLI or SDK call. Never guess the region or account ID — always use this skill to read it from the local configuration first.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[809,810,811],{"name":22,"slug":23,"type":14},{"name":794,"slug":795,"type":14},{"name":812,"slug":813,"type":14},"Configuration","configuration","2026-07-08T05:55:33.716099",{"slug":112,"name":112,"fn":816,"description":817,"org":818,"tags":819,"stars":24,"repoUrl":25,"updatedAt":828},"set up Python environments for AWS","Set up an isolated Python environment for SageMaker \u002F AWS work, with the right Python version and current boto3. Use this skill whenever Python code will be executed for a SageMaker deployment, training job, or any AWS automation — including when about to run `pip install`, when about to invoke `boto3`, when creating or activating a virtualenv, or when the user asks to \"set up the environment\". Never use system Python and never `pip install` into it. Always isolate. This skill prevents the most common failure modes: wrong Python version, dependency conflicts, and stale SDKs.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[820,821,822,825],{"name":19,"slug":20,"type":14},{"name":22,"slug":23,"type":14},{"name":823,"slug":824,"type":14},"Engineering","engineering",{"name":826,"slug":827,"type":14},"Python","python","2026-07-08T05:55:32.505017",{"slug":4,"name":4,"fn":5,"description":6,"org":830,"tags":831,"stars":24,"repoUrl":25,"updatedAt":26},{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[832,833,834,835],{"name":19,"slug":20,"type":14},{"name":22,"slug":23,"type":14},{"name":16,"slug":17,"type":14},{"name":9,"slug":13,"type":14},{"slug":127,"name":127,"fn":837,"description":838,"org":839,"tags":840,"stars":24,"repoUrl":25,"updatedAt":849},"configure SageMaker IAM roles","Ensure a usable SageMaker execution role exists before deploying or training. Use this skill whenever about to create a SageMaker endpoint, model, training job, or any resource that requires an execution role. Use it especially when the user has not provided a role ARN explicitly, when scripts are about to call `iam:CreateRole`, or when an AccessDenied error mentions an IAM action. Never blindly call `iam:CreateRole` — always check for existing roles first. This skill prevents the most common SageMaker deployment failure: trying to create IAM resources from an SSO principal that has no IAM write permissions.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[841,842,843,846],{"name":19,"slug":20,"type":14},{"name":22,"slug":23,"type":14},{"name":844,"slug":845,"type":14},"Permissions","permissions",{"name":847,"slug":848,"type":14},"Security","security","2026-07-08T05:55:34.948771",{"slug":156,"name":156,"fn":851,"description":852,"org":853,"tags":854,"stars":24,"repoUrl":25,"updatedAt":862},"create production-ready SageMaker endpoints","Create a SageMaker endpoint (real-time or async) with autoscaling, CloudWatch alarms, and tagging enabled by default. Use this skill whenever about to create a SageMaker endpoint, write deployment code that calls `create_endpoint`, or finalize a deployment after the image URI and IAM role are known. Provides deploy.py for real-time endpoints and deploy_async.py for async endpoints (with genuine scale-to-zero support). This is the last step in the SageMaker deployment workflow. Never generate a bare `create_endpoint` call without these defaults — endpoints without autoscaling or alarms are demos, not deployments.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[855,856,857,858,859],{"name":19,"slug":20,"type":14},{"name":22,"slug":23,"type":14},{"name":16,"slug":17,"type":14},{"name":9,"slug":13,"type":14},{"name":860,"slug":861,"type":14},"Monitoring","monitoring","2026-07-08T05:55:38.664702",{"slug":142,"name":142,"fn":864,"description":865,"org":866,"tags":867,"stars":24,"repoUrl":25,"updatedAt":871},"select SageMaker serving containers","Pick the right serving container for a SageMaker model deployment and find its current image URI. Use this skill whenever about to deploy a model to a SageMaker endpoint and an image URI needs to be chosen — including when the user says \"deploy this LLM\", \"host this HuggingFace model\", \"serve this fine-tuned model\", \"deploy this embedding model\", \"host a reranker\", \"serve a sentence-transformers model\", or when about to hardcode any container URI in deployment code. HuggingFace-curated Deep Learning Containers are ALWAYS preferred: HuggingFace vLLM (LLMs and generative rerankers), HuggingFace vLLM-Omni (multimodal), TEI (embeddings\u002Fcross-encoder rerankers), HF Inference Toolkit (other transformers). Generic images (AWS vLLM, DJL-LMI, SGLang) are used only when no HuggingFace image is compatible — never merely because they carry a newer version. Never hardcode a container URI from memory and never default to TGI. Prevents stale-image failures and wrong-region URIs.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[868,869,870],{"name":19,"slug":20,"type":14},{"name":22,"slug":23,"type":14},{"name":16,"slug":17,"type":14},"2026-07-08T05:55:36.173465",24,{"items":874,"total":997},[875,893,907,914,920,927,934,941,949,955,968,982],{"slug":876,"name":876,"fn":877,"description":878,"org":879,"tags":880,"stars":890,"repoUrl":891,"updatedAt":892},"train-sentence-transformers","train sentence-transformers models","Train or fine-tune sentence-transformers models across `SentenceTransformer` (bi-encoder; dense or static embedding model; for retrieval, similarity, clustering, classification, paraphrase mining, dedup, multimodal), `CrossEncoder` (reranker; pair scoring for two-stage retrieval \u002F pair classification), and `SparseEncoder` (SPLADE, sparse embedding model; for learned-sparse retrieval). Covers loss selection, hard-negative mining, evaluators, distillation, LoRA, Matryoshka, and Hugging Face Hub publishing. Use for any sentence-transformers training task.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[881,884,885,886,887],{"name":882,"slug":883,"type":14},"Deep Learning","deep-learning",{"name":9,"slug":13,"type":14},{"name":801,"slug":802,"type":14},{"name":826,"slug":827,"type":14},{"name":888,"slug":889,"type":14},"Search","search",18914,"https:\u002F\u002Fgithub.com\u002Fhuggingface\u002Fsentence-transformers","2026-05-08T05:09:16.820066",{"slug":894,"name":894,"fn":895,"description":896,"org":897,"tags":898,"stars":904,"repoUrl":905,"updatedAt":906},"trl-training","train and fine-tune LLMs with TRL","Train and fine-tune transformer language models using TRL (Transformers Reinforcement Learning). Supports SFT, DPO, GRPO, KTO, RLOO and Reward Model training via CLI commands.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[899,900,901,902,903],{"name":19,"slug":20,"type":14},{"name":882,"slug":883,"type":14},{"name":9,"slug":13,"type":14},{"name":801,"slug":802,"type":14},{"name":826,"slug":827,"type":14},18850,"https:\u002F\u002Fgithub.com\u002Fhuggingface\u002Ftrl","2026-04-06T18:25:32.746828",{"slug":788,"name":788,"fn":789,"description":790,"org":908,"tags":909,"stars":24,"repoUrl":25,"updatedAt":803},{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[910,911,912,913],{"name":794,"slug":795,"type":14},{"name":797,"slug":798,"type":14},{"name":9,"slug":13,"type":14},{"name":801,"slug":802,"type":14},{"slug":104,"name":104,"fn":805,"description":806,"org":915,"tags":916,"stars":24,"repoUrl":25,"updatedAt":814},{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[917,918,919],{"name":22,"slug":23,"type":14},{"name":794,"slug":795,"type":14},{"name":812,"slug":813,"type":14},{"slug":112,"name":112,"fn":816,"description":817,"org":921,"tags":922,"stars":24,"repoUrl":25,"updatedAt":828},{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[923,924,925,926],{"name":19,"slug":20,"type":14},{"name":22,"slug":23,"type":14},{"name":823,"slug":824,"type":14},{"name":826,"slug":827,"type":14},{"slug":4,"name":4,"fn":5,"description":6,"org":928,"tags":929,"stars":24,"repoUrl":25,"updatedAt":26},{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[930,931,932,933],{"name":19,"slug":20,"type":14},{"name":22,"slug":23,"type":14},{"name":16,"slug":17,"type":14},{"name":9,"slug":13,"type":14},{"slug":127,"name":127,"fn":837,"description":838,"org":935,"tags":936,"stars":24,"repoUrl":25,"updatedAt":849},{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[937,938,939,940],{"name":19,"slug":20,"type":14},{"name":22,"slug":23,"type":14},{"name":844,"slug":845,"type":14},{"name":847,"slug":848,"type":14},{"slug":156,"name":156,"fn":851,"description":852,"org":942,"tags":943,"stars":24,"repoUrl":25,"updatedAt":862},{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[944,945,946,947,948],{"name":19,"slug":20,"type":14},{"name":22,"slug":23,"type":14},{"name":16,"slug":17,"type":14},{"name":9,"slug":13,"type":14},{"name":860,"slug":861,"type":14},{"slug":142,"name":142,"fn":864,"description":865,"org":950,"tags":951,"stars":24,"repoUrl":25,"updatedAt":871},{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[952,953,954],{"name":19,"slug":20,"type":14},{"name":22,"slug":23,"type":14},{"name":16,"slug":17,"type":14},{"slug":956,"name":956,"fn":957,"description":958,"org":959,"tags":960,"stars":24,"repoUrl":25,"updatedAt":967},"hf-mcp","access Hugging Face Hub via MCP","Use Hugging Face Hub via MCP server tools. Search models, datasets, Spaces, papers. Get repo details, fetch documentation, run compute jobs, and use Gradio Spaces as AI tools. Available when connected to the HF MCP server.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[961,962,963,964],{"name":797,"slug":798,"type":14},{"name":9,"slug":13,"type":14},{"name":801,"slug":802,"type":14},{"name":965,"slug":966,"type":14},"MCP","mcp","2026-04-06T18:25:50.364185",{"slug":969,"name":969,"fn":970,"description":971,"org":972,"tags":973,"stars":24,"repoUrl":25,"updatedAt":981},"hf-mem","estimate memory for Hugging Face models","Hugging Face CLI to estimate the required memory to load Safetensors or GGUF model weights for inference from the Hugging Face Hub",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[974,975,976,977,978],{"name":19,"slug":20,"type":14},{"name":794,"slug":795,"type":14},{"name":9,"slug":13,"type":14},{"name":801,"slug":802,"type":14},{"name":979,"slug":980,"type":14},"Performance","performance","2026-06-13T07:23:57.101435",{"slug":983,"name":983,"fn":984,"description":985,"org":986,"tags":987,"stars":24,"repoUrl":25,"updatedAt":996},"huggingface-best","find and compare Hugging Face models","Use when the user asks about finding the best, top, or recommended model for a task, wants to know what AI model to use, or wants to compare models by benchmark scores. Triggers on: \"best model for X\", \"what model should I use for\", \"top models for [task]\", \"which model runs on my laptop\u002Fmachine\u002Fdevice\", \"recommend a model for\", \"what LLM should I use for\", \"compare models for\", \"what's state of the art for\", or any question about choosing an AI model for a specific use case. Always use this skill when the user wants model recommendations or comparisons, even if they don't explicitly mention HuggingFace or benchmarks.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[988,991,992,993],{"name":989,"slug":990,"type":14},"Analytics","analytics",{"name":9,"slug":13,"type":14},{"name":801,"slug":802,"type":14},{"name":994,"slug":995,"type":14},"Research","research","2026-04-24T05:09:45.870658",37]