[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7EN71XDDTTJEXK70WRTT83B":3,"skill-same:01M7EN71XDDTTJEXK70WRTT83B":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},"01M7EN71XDDTTJEXK70WRTT83B","groq-guide",null,"Ultra-low-latency inference on Groq's LPU hardware — when tokens-per-second is the product requirement.","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-groq-guide\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: groq-guide\ndescription: Ultra-low-latency inference on Groq's LPU hardware — when tokens-per-second is the product requirement.\ncategory: ai-research\n---\n\n## Overview\n\nGroq runs language models on its own custom silicon — Language Processing Units\n(LPUs) — delivering exceptionally high tokens-per-second and low latency for\nsupported open models. The value proposition is simple and sharp: when speed is\nthe binding constraint, Groq's hardware-tuned inference is among the fastest\navailable via API.\n\nThe tradeoff is model selection: Groq supports a curated set of models compiled\nfor its hardware, not the full open-model universe. You trade breadth for\nvelocity. For applications where latency defines the experience — voice agents,\nreal-time copilots, interactive tools, agents making many sequential calls —\nthat trade is often worth it.\n\nUse Groq when you've measured your latency requirements and they demand\nhardware-level speed. It's a specialist tool: unmatched at its specialty,\nnarrow outside it.\n\n## When to use\n\n- Real-time applications: voice agents, live copilots, interactive chat where\n  latency is the UX.\n- Agent loops with many sequential model calls — per-call latency compounds.\n- High-throughput batch processing where tokens\u002Fsec per dollar wins.\n- Demos and prototypes where responsiveness sells the experience.\n- Any workload where you can use a Groq-supported model and speed is the top\n  requirement.\n- Latency-sensitive evals and benchmarking.\n\n## Core concepts\n\n- **LPU hardware**: purpose-built chips for LLM inference — deterministic,\n  high-throughput execution. The speed comes from hardware\u002Fsoftware co-design,\n  not just optimization tricks.\n- **Curated model set**: models compiled and supported for LPU execution. Check\n  current availability — the set grows over time but remains curated, not\n  comprehensive.\n- **Tokens-per-second leadership**: the headline metric — Groq typically leads\n  on raw generation speed for supported models. Validate on your prompt shapes.\n- **Low time-to-first-token**: fast first tokens matter for interactive UX;\n  Groq's architecture delivers here too. Measure TTFT for your use case.\n- **API compatibility**: OpenAI-compatible API surface — swapping in Groq is\n  usually a base-URL and key change, making evaluation cheap.\n- **Rate limits and tiers**: generous free tiers for development; paid tiers for\n  production throughput. Understand the rate limits for your expected concurrency.\n- **Deterministic performance**: LPU execution is highly deterministic — less\n  latency variance than GPU-based serving. Useful for SLO planning.\n- **Model versioning**: supported models update over time; pin what's\n  pinnable and re-validate on changes.\n\n## Practical workflow\n\n1. **Confirm model availability.** Check that a model meeting your quality bar\n   is supported on Groq. If none is, Groq's speed doesn't help — quality bar\n   first.\n2. **Benchmark your workload.** Measure TTFT, tokens\u002Fsec, and end-to-end latency\n   on your actual prompts. Confirm the speed advantage holds for your shapes\n   (long contexts, structured outputs).\n3. **Evaluate quality parity.** Run your eval set — ensure the Groq-served model\n   meets your quality bar. Hardware-optimized serving should be behavior-preserving;\n   verify, don't assume.\n4. **Design for the speed.** Ultra-low latency enables new UX: streaming voice,\n   real-time collaboration, rapid agent loops. Don't just port a slow design —\n   exploit the speed.\n5. **Plan rate limits.** Map your production concurrency to Groq's rate limits\n   and tiers. Load-test at expected peak; arrange capacity before launch.\n6. **Build fallbacks.** For production, configure fallback providers — speed\n   specialists are still single points of failure.\n7. **Monitor latency in production.** Track TTFT and tokens\u002Fsec on live traffic;\n   alert on regressions. Deterministic hardware should give stable numbers —\n   instability signals a problem.\n\nChecklist for Groq in production:\n- Supported model clears your quality bar on your evals.\n- Latency benchmarked on your prompts (TTFT + throughput).\n- Rate limits mapped to production concurrency.\n- Fallback providers configured.\n- Production latency monitored with alerts.\n\n## Common pitfalls\n\n- **Speed without a quality bar.** Choosing Groq's fastest model instead of the\n  slowest model that meets your quality needs. Quality gates the choice; speed\n  ranks among qualifiers.\n- **Model set mismatch.** Designing around a model Groq doesn't support, then\n  compromising quality to get the speed. Check availability first.\n- **Ignoring rate limits.** Free-tier limits in development masking production\n  needs. Plan capacity for launch traffic, not dev traffic.\n- **No fallback.** Groq as a single point of failure for a latency-critical\n  product. Incidents happen to everyone.\n- **Over-engineering for speed you don't need.** Batch\u002Foffline workloads paying\n  a premium (in model choice constraints) for latency they can't use. Match the\n  tool to the requirement.\n- **Assuming quality parity.** Hardware-optimized inference is usually\n  behavior-preserving, but \"usually\" isn't a guarantee — run your evals.\n- **Long-context surprises.** Speed advantages can narrow at very long contexts.\n  Benchmark at your actual context lengths.\n- **Latency SLOs without monitoring.** Assuming deterministic hardware means no\n  monitoring needed. Monitor anyway — networks, queues, and rate limits still\n  vary.\n","## Overview\n\nGroq runs language models on its own custom silicon — Language Processing Units\n(LPUs) — delivering exceptionally high tokens-per-second and low latency for\nsupported open models. The value proposition is simple and sharp: when speed is\nthe binding constraint, Groq's hardware-tuned inference is among the fastest\navailable via API.\n\nThe tradeoff is model selection: Groq supports a curated set of models compiled\nfor its hardware, not the full open-model universe. You trade breadth for\nvelocity. For applications where latency defines the experience — voice agents,\nreal-time copilots, interactive tools, agents making many sequential calls —\nthat trade is often worth it.\n\nUse Groq when you've measured your latency requirements and they demand\nhardware-level speed. It's a specialist tool: unmatched at its specialty,\nnarrow outside it.\n\n## When to use\n\n- Real-time applications: voice agents, live copilots, interactive chat where\n  latency is the UX.\n- Agent loops with many sequential model calls — per-call latency compounds.\n- High-throughput batch processing where tokens\u002Fsec per dollar wins.\n- Demos and prototypes where responsiveness sells the experience.\n- Any workload where you can use a Groq-supported model and speed is the top\n  requirement.\n- Latency-sensitive evals and benchmarking.\n\n## Core concepts\n\n- **LPU hardware**: purpose-built chips for LLM inference — deterministic,\n  high-throughput execution. The speed comes from hardware\u002Fsoftware co-design,\n  not just optimization tricks.\n- **Curated model set**: models compiled and supported for LPU execution. Check\n  current availability — the set grows over time but remains curated, not\n  comprehensive.\n- **Tokens-per-second leadership**: the headline metric — Groq typically leads\n  on raw generation speed for supported models. Validate on your prompt shapes.\n- **Low time-to-first-token**: fast first tokens matter for interactive UX;\n  Groq's architecture delivers here too. Measure TTFT for your use case.\n- **API compatibility**: OpenAI-compatible API surface — swapping in Groq is\n  usually a base-URL and key change, making evaluation cheap.\n- **Rate limits and tiers**: generous free tiers for development; paid tiers for\n  production throughput. Understand the rate limits for your expected concurrency.\n- **Deterministic performance**: LPU execution is highly deterministic — less\n  latency variance than GPU-based serving. Useful for SLO planning.\n- **Model versioning**: supported models update over time; pin what's\n  pinnable and re-validate on changes.\n\n## Practical workflow\n\n1. **Confirm model availability.** Check that a model meeting your quality bar\n   is supported on Groq. If none is, Groq's speed doesn't help — quality bar\n   first.\n2. **Benchmark your workload.** Measure TTFT, tokens\u002Fsec, and end-to-end latency\n   on your actual prompts. Confirm the speed advantage holds for your shapes\n   (long contexts, structured outputs).\n3. **Evaluate quality parity.** Run your eval set — ensure the Groq-served model\n   meets your quality bar. Hardware-optimized serving should be behavior-preserving;\n   verify, don't assume.\n4. **Design for the speed.** Ultra-low latency enables new UX: streaming voice,\n   real-time collaboration, rapid agent loops. Don't just port a slow design —\n   exploit the speed.\n5. **Plan rate limits.** Map your production concurrency to Groq's rate limits\n   and tiers. Load-test at expected peak; arrange capacity before launch.\n6. **Build fallbacks.** For production, configure fallback providers — speed\n   specialists are still single points of failure.\n7. **Monitor latency in production.** Track TTFT and tokens\u002Fsec on live traffic;\n   alert on regressions. Deterministic hardware should give stable numbers —\n   instability signals a problem.\n\nChecklist for Groq in production:\n- Supported model clears your quality bar on your evals.\n- Latency benchmarked on your prompts (TTFT + throughput).\n- Rate limits mapped to production concurrency.\n- Fallback providers configured.\n- Production latency monitored with alerts.\n\n## Common pitfalls\n\n- **Speed without a quality bar.** Choosing Groq's fastest model instead of the\n  slowest model that meets your quality needs. Quality gates the choice; speed\n  ranks among qualifiers.\n- **Model set mismatch.** Designing around a model Groq doesn't support, then\n  compromising quality to get the speed. Check availability first.\n- **Ignoring rate limits.** Free-tier limits in development masking production\n  needs. Plan capacity for launch traffic, not dev traffic.\n- **No fallback.** Groq as a single point of failure for a latency-critical\n  product. Incidents happen to everyone.\n- **Over-engineering for speed you don't need.** Batch\u002Foffline workloads paying\n  a premium (in model choice constraints) for latency they can't use. Match the\n  tool to the requirement.\n- **Assuming quality parity.** Hardware-optimized inference is usually\n  behavior-preserving, but \"usually\" isn't a guarantee — run your evals.\n- **Long-context surprises.** Speed advantages can narrow at very long contexts.\n  Benchmark at your actual context lengths.\n- **Latency SLOs without monitoring.** Assuming deterministic hardware means no\n  monitoring needed. Monitor anyway — networks, queues, and rate limits still\n  vary.\n",{"name":5,"category":24,"description":7},"ai-research","main","af388a77c2fe5e0b4fd879cca4300a0ffa9a6579","skills\u002Fams-groq-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 groq-guide",{"tool":39,"command":40},"bunx","bunx skills add TanNhatCMS\u002Fjarvis --skill groq-guide",{"tool":42,"command":43},"pnpm","pnpm dlx skills add TanNhatCMS\u002Fjarvis --skill groq-guide","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fams-groq-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",[],[]]