[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7EMA8TAEWFJB8CD5K82HGSD":3,"skill-same:01M7EMA8TAEWFJB8CD5K82HGSD":47},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":25,"commitSha":26,"skillDirectory":27,"forksCount":28,"repositoryHtmlUrl":29,"repositoryHomepageUrl":6,"ownerHtmlUrl":30,"ownerBlogUrl":31,"ownerDescription":6,"updatedAt":32,"install":33,"resources":46},"01M7EMA8TAEWFJB8CD5K82HGSD","anyscale-guide",null,"Scale Ray workloads and serve models on Anyscale — managed Ray clusters for training and inference.","TanNhatCMS\u002Fjarvis","TanNhatCMS","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F113796420?v=4","01M753MV17KG5CD3152G3WQEQY","KietNT","01M7EKEDYVYQ6PWVJJMZ911VMX",0,"MIT","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fams-anyscale-guide\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: anyscale-guide\ndescription: Scale Ray workloads and serve models on Anyscale — managed Ray clusters for training and inference.\ncategory: ai-research\n---\n\n## Overview\n\nAnyscale is the managed platform for Ray, the distributed computing framework —\nit runs your Ray workloads (training, batch processing, hyperparameter tuning,\nmodel serving with Ray Serve) on managed clusters without you operating the\ninfrastructure. For AI teams, the practical shape is: develop Ray code locally,\nthen run it at scale on Anyscale with managed cluster lifecycle, observability,\nand cost controls.\n\nThe distinctive value is Ray itself: a unified programming model for distributed\nPython that covers training, serving, and data processing in one framework. If\nyour ML work spans \"train a model,\" \"tune hyperparameters,\" \"run batch\ninference,\" and \"serve the model,\" Ray does all four — and Anyscale removes the\ncluster-management burden. It's infrastructure for teams whose workloads are\ngenuinely distributed, not just \"a GPU in the cloud.\"\n\nAnyscale's lens: if you're reaching for Ray, use the managed platform; if you're\nnot using Ray, Anyscale probably isn't your starting point.\n\n## When to use\n\n- Distributed training jobs that need multi-node\u002Fmulti-GPU orchestration.\n- Hyperparameter sweeps at scale (Ray Tune) without managing the cluster.\n- Model serving with Ray Serve: composable deployments, multi-model pipelines.\n- Batch inference over large datasets with distributed processing.\n- Reinforcement learning workloads (Ray RLlib) at scale.\n- Teams standardizing on Ray as their distributed compute layer.\n\n## Core concepts\n\n- **Ray core primitives**: tasks (stateless functions), actors (stateful\n  workers), and objects (distributed shared memory). Learn these three — every\n  Ray program composes them.\n- **Managed clusters**: Anyscale provisions and manages the cluster lifecycle\n  (autoscaling node pools, spot instances, upgrades). You declare compute\n  config; the platform handles the rest.\n- **Ray Tune**: distributed hyperparameter optimization with schedulers\n  (ASHA\u002FPBT) that kill bad trials early. The cost savings from early stopping\n  usually dwarf the platform cost.\n- **Ray Serve**: model serving with deployment graphs — compose models,\n  preprocessors, and business logic into one served application with\n  autoscaling per deployment.\n- **Workspaces**: interactive development environments on the cluster — develop\n  against real distributed resources, not just your laptop.\n- **Jobs**: the production unit — submit a Ray job (training run, batch\n  pipeline) with pinned code, dependencies, and compute config. Jobs are\n  auditable and reproducible.\n- **Autoscaling and spot**: clusters scale node pools with demand; spot\n  instances cut cost for fault-tolerant workloads. Configure per workload —\n  serving on spot needs care.\n- **Observability**: Ray dashboard plus Anyscale's monitoring — task timelines,\n  actor states, GPU utilization, logs. Distributed debugging lives here.\n\n## Practical workflow\n\n1. **Develop the Ray program locally.** Write tasks\u002Factors\u002Fserve deployments\n   against a local Ray cluster. Get the logic right at small scale first.\n2. **Define compute config.** Node types, autoscaling bounds, GPU requirements\n   per workload. Start modest; scale based on measured utilization.\n3. **Run as a Job for anything repeatable.** Notebooks and workspaces for\n   exploration; Jobs for training runs and pipelines — reproducible, logged,\n   reviewable.\n4. **Tune with early stopping.** For hyperparameter search, use ASHA or PBT\n   schedulers — killing bad trials early is the highest-ROI optimization in\n   distributed tuning.\n5. **Serve with Ray Serve graphs.** Compose the inference pipeline (preprocess →\n   model → postprocess) as a deployment graph with per-deployment scaling.\n   Load-test the graph, not just the model.\n6. **Use spot where fault-tolerant.** Training with checkpointing and batch\n   jobs tolerate preemption — spot instances cut those costs substantially.\n   Keep serving and stateful work on on-demand.\n7. **Monitor and right-size.** Watch GPU utilization, autoscaling behavior, and\n   cost per job. Distributed systems drift toward waste — review regularly.\n\nChecklist for Anyscale production work:\n- Jobs pinned (code, deps, compute config) and reproducible.\n- Autoscaling bounds set per workload; spot policy deliberate.\n- Tuning uses early-stopping schedulers.\n- Serve graphs load-tested end-to-end.\n- GPU utilization and cost-per-job monitored.\n\n## Common pitfalls\n\n- **Ray for non-distributed work.** A single-GPU training job doesn't need Ray's\n  machinery. Use Ray when the workload is genuinely distributed.\n- **No checkpointing with spot.** Spot preemptions without checkpoints waste\n  the entire run. Checkpoint frequently; make training resumable.\n- **Autoscaling misconfiguration.** Bounds too tight (jobs queue forever) or too\n  loose (idle GPUs burning money). Tune from utilization data.\n- **Tuning without early stopping.** Full-fidelity training of every\n  hyperparameter combination. ASHA\u002FPBT exist precisely to avoid this.\n- **Serving the model, not the graph.** Load-testing the bare model while\n  production runs a preprocess→model→postprocess graph with different scaling\n  needs per stage.\n- **Workspace-as-production.** Running production workloads from interactive\n  workspaces instead of Jobs. Workspaces are for development; Jobs are for\n  repeatability.\n- **Ignoring data locality.** Distributed compute with centralized data access\n  bottlenecks on I\u002FO. Co-locate data access with compute; use Ray Data\n  thoughtfully.\n- **Cost blindness.** Multi-node GPU clusters without per-job cost tracking.\n  Distributed waste scales as fast as distributed compute.\n","## Overview\n\nAnyscale is the managed platform for Ray, the distributed computing framework —\nit runs your Ray workloads (training, batch processing, hyperparameter tuning,\nmodel serving with Ray Serve) on managed clusters without you operating the\ninfrastructure. For AI teams, the practical shape is: develop Ray code locally,\nthen run it at scale on Anyscale with managed cluster lifecycle, observability,\nand cost controls.\n\nThe distinctive value is Ray itself: a unified programming model for distributed\nPython that covers training, serving, and data processing in one framework. If\nyour ML work spans \"train a model,\" \"tune hyperparameters,\" \"run batch\ninference,\" and \"serve the model,\" Ray does all four — and Anyscale removes the\ncluster-management burden. It's infrastructure for teams whose workloads are\ngenuinely distributed, not just \"a GPU in the cloud.\"\n\nAnyscale's lens: if you're reaching for Ray, use the managed platform; if you're\nnot using Ray, Anyscale probably isn't your starting point.\n\n## When to use\n\n- Distributed training jobs that need multi-node\u002Fmulti-GPU orchestration.\n- Hyperparameter sweeps at scale (Ray Tune) without managing the cluster.\n- Model serving with Ray Serve: composable deployments, multi-model pipelines.\n- Batch inference over large datasets with distributed processing.\n- Reinforcement learning workloads (Ray RLlib) at scale.\n- Teams standardizing on Ray as their distributed compute layer.\n\n## Core concepts\n\n- **Ray core primitives**: tasks (stateless functions), actors (stateful\n  workers), and objects (distributed shared memory). Learn these three — every\n  Ray program composes them.\n- **Managed clusters**: Anyscale provisions and manages the cluster lifecycle\n  (autoscaling node pools, spot instances, upgrades). You declare compute\n  config; the platform handles the rest.\n- **Ray Tune**: distributed hyperparameter optimization with schedulers\n  (ASHA\u002FPBT) that kill bad trials early. The cost savings from early stopping\n  usually dwarf the platform cost.\n- **Ray Serve**: model serving with deployment graphs — compose models,\n  preprocessors, and business logic into one served application with\n  autoscaling per deployment.\n- **Workspaces**: interactive development environments on the cluster — develop\n  against real distributed resources, not just your laptop.\n- **Jobs**: the production unit — submit a Ray job (training run, batch\n  pipeline) with pinned code, dependencies, and compute config. Jobs are\n  auditable and reproducible.\n- **Autoscaling and spot**: clusters scale node pools with demand; spot\n  instances cut cost for fault-tolerant workloads. Configure per workload —\n  serving on spot needs care.\n- **Observability**: Ray dashboard plus Anyscale's monitoring — task timelines,\n  actor states, GPU utilization, logs. Distributed debugging lives here.\n\n## Practical workflow\n\n1. **Develop the Ray program locally.** Write tasks\u002Factors\u002Fserve deployments\n   against a local Ray cluster. Get the logic right at small scale first.\n2. **Define compute config.** Node types, autoscaling bounds, GPU requirements\n   per workload. Start modest; scale based on measured utilization.\n3. **Run as a Job for anything repeatable.** Notebooks and workspaces for\n   exploration; Jobs for training runs and pipelines — reproducible, logged,\n   reviewable.\n4. **Tune with early stopping.** For hyperparameter search, use ASHA or PBT\n   schedulers — killing bad trials early is the highest-ROI optimization in\n   distributed tuning.\n5. **Serve with Ray Serve graphs.** Compose the inference pipeline (preprocess →\n   model → postprocess) as a deployment graph with per-deployment scaling.\n   Load-test the graph, not just the model.\n6. **Use spot where fault-tolerant.** Training with checkpointing and batch\n   jobs tolerate preemption — spot instances cut those costs substantially.\n   Keep serving and stateful work on on-demand.\n7. **Monitor and right-size.** Watch GPU utilization, autoscaling behavior, and\n   cost per job. Distributed systems drift toward waste — review regularly.\n\nChecklist for Anyscale production work:\n- Jobs pinned (code, deps, compute config) and reproducible.\n- Autoscaling bounds set per workload; spot policy deliberate.\n- Tuning uses early-stopping schedulers.\n- Serve graphs load-tested end-to-end.\n- GPU utilization and cost-per-job monitored.\n\n## Common pitfalls\n\n- **Ray for non-distributed work.** A single-GPU training job doesn't need Ray's\n  machinery. Use Ray when the workload is genuinely distributed.\n- **No checkpointing with spot.** Spot preemptions without checkpoints waste\n  the entire run. Checkpoint frequently; make training resumable.\n- **Autoscaling misconfiguration.** Bounds too tight (jobs queue forever) or too\n  loose (idle GPUs burning money). Tune from utilization data.\n- **Tuning without early stopping.** Full-fidelity training of every\n  hyperparameter combination. ASHA\u002FPBT exist precisely to avoid this.\n- **Serving the model, not the graph.** Load-testing the bare model while\n  production runs a preprocess→model→postprocess graph with different scaling\n  needs per stage.\n- **Workspace-as-production.** Running production workloads from interactive\n  workspaces instead of Jobs. Workspaces are for development; Jobs are for\n  repeatability.\n- **Ignoring data locality.** Distributed compute with centralized data access\n  bottlenecks on I\u002FO. Co-locate data access with compute; use Ray Data\n  thoughtfully.\n- **Cost blindness.** Multi-node GPU clusters without per-job cost tracking.\n  Distributed waste scales as fast as distributed compute.\n",{"name":5,"category":24,"description":7},"ai-research","main","af388a77c2fe5e0b4fd879cca4300a0ffa9a6579","skills\u002Fams-anyscale-guide",1,"https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis","https:\u002F\u002Fgithub.com\u002FTanNhatCMS","https:\u002F\u002Fmrkiet.name.vn","2026-11-07 14:23:08",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":45},[35,38,41],{"tool":36,"command":37},"npx","npx skills add TanNhatCMS\u002Fjarvis --skill anyscale-guide",{"tool":39,"command":40},"bunx","bunx skills add TanNhatCMS\u002Fjarvis --skill anyscale-guide",{"tool":42,"command":43},"pnpm","pnpm dlx skills add TanNhatCMS\u002Fjarvis --skill anyscale-guide","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fams-anyscale-guide\u002FSKILL.md",true,[],[48,55,63,70,77,84],{"skillNo":49,"name":50,"nameZh":6,"description":51,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":52,"categories":53,"keywords":54,"publishedAt":19},"01M7EWYJR0J8DARY7E59Z6Y9MV","web-scrape","Turn web pages into LLM-ready markdown via the Firecrawl API (renders JS, anti-bot handling). Use when browser_open\u002Ffetch fails on JS-heavy pages, or when you need clean markdown\u002Fstructured extraction from a site. Requires FIRECRAWL_API_KEY.","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fweb-scrape\u002FSKILL.md",[],[],{"skillNo":56,"name":57,"nameZh":6,"description":58,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":59,"sourceHtmlUrl":60,"categories":61,"keywords":62,"publishedAt":19},"01M7EWYBN5P6F2AJG3FR24T9BP","design-taste-frontend","tasteskill v2, the anti-slop frontend skill. Art direction for websites, distilled from a study of 247 curated gallery sites (One Page Love, Awwwards Sites of the Day). Deals researched directions (idea, composition, type, palette, photographs, section blueprints) from a library, enforces a strict no-slop house style with a real-browser inspector, and runs a look-and-fix loop on screenshots. Use this whenever you build, design, redesign, restyle, polish or review a landing page, marketing site, portfolio, product page, one-pager, hero or any public-facing web page, in any stack (HTML\u002FCSS, React, Next.js, Vue, Svelte, Astro, Tailwind) — also when the user says a page looks generic, templated, \"AI-made\", sloppy, or asks to make it look designed, premium, award-level or less like every other site. Not for dashboards or app screens.","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Ftaste-design\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Ftaste-design\u002FSKILL.md",[],[],{"skillNo":64,"name":65,"nameZh":6,"description":66,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":67,"categories":68,"keywords":69,"publishedAt":19},"01M7EWY8FQFNQ2GB7NZZWGSSDX","test-driven-development","Use when implementing any feature or bugfix, before writing implementation code","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fsuperpowers-tdd\u002FSKILL.md",[],[],{"skillNo":71,"name":72,"nameZh":6,"description":73,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":19},"01M7EWY5SHBFRH1HDHVWM9K7TV","systematic-debugging","Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fsuperpowers-debugging\u002FSKILL.md",[],[],{"skillNo":78,"name":79,"nameZh":6,"description":80,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":81,"categories":82,"keywords":83,"publishedAt":19},"01M7EWY3NM0Z0JAKS7HDZ24K66","requesting-code-review","Use when completing tasks, implementing major features, or before merging to verify work meets requirements","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fsuperpowers-code-review\u002FSKILL.md",[],[],{"skillNo":85,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":90,"publishedAt":19},"01M7EWY0WRMWX4M9ZFY0WKC593","brainstorming","You MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).","https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fsuperpowers-brainstorming\u002FSKILL.md",[],[]]