[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7ESXDY305PAMKJD5RXDJ232":3,"skill-same:01M7ESXDY305PAMKJD5RXDJ232":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},"01M7ESXDY305PAMKJD5RXDJ232","llamacpp-guide",null,"Run and optimize models with llama.cpp — GGUF quantization, CPU\u002FGPU offloading, server mode, and performance tuning. Use when you need maximum local inference efficiency or cross-platform deployment.","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-llamacpp-guide\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: llamacpp-guide\ndescription: Run and optimize models with llama.cpp — GGUF quantization, CPU\u002FGPU offloading, server mode, and performance tuning. Use when you need maximum local inference efficiency or cross-platform deployment.\ncategory: ai-research\n---\n\n# llama.cpp Guide\n\nllama.cpp is the portable inference engine: C++-based, runs everywhere from phones to servers, with \nthe best CPU inference in the business and broad GGUF model support. It's the power tool behind \nmost serious local LLM setups.\n\n## Overview\n\nThe mental model: models in GGUF format (quantized variants from 2-bit to 8-bit), an engine that \nsplits work between CPU and GPU layers, and two interfaces — a CLI for direct use and a server \nmode exposing an OpenAI-compatible API. Performance comes from tuning thread counts, batch sizes, \nand GPU offloading to your specific hardware. It's lower-level than packaged runners, which means \nmore control and more knobs.\n\n## When to use\n\n- Maximum local performance: squeezing the most tokens\u002Fsec from your hardware.\n- CPU-only inference where other engines struggle.\n- Cross-platform deployment: same engine on Mac, Linux, Windows, even mobile.\n- Fine-grained control: custom quantization, speculative decoding, grammar-constrained output.\n\n## Core concepts\n\n- **GGUF format**: the model container — architecture, weights, and metadata in one file, with \nquantization variants (Q4_K_M is the popular sweet spot). Choose the largest quant that fits your \nRAM.\n- **GPU offloading**: layers assigned to GPU (`-ngl`) vs. CPU. Full offload when VRAM fits; partial \noffload still beats CPU-only. More offloaded layers = faster.\n- **Threading and batching**: CPU thread count matched to your cores (physical, not hyperthreaded, \nusually); batch size tuned for prompt processing vs. generation.\n- **Server mode**: HTTP server with OpenAI-compatible endpoints — the integration point for apps \nand agents. Supports concurrent requests with continuous batching.\n- **Context management**: context size set at load (`-c`); KV cache quantization to stretch long \ncontexts into limited RAM.\n- **Advanced features**: speculative decoding (draft model proposes, main model verifies — big \nspeedups), grammar-constrained generation (JSON mode via grammars), multimodal support.\n\n## Practical workflow\n\n1. Pick the GGUF quant: largest that fits comfortably in RAM (leave headroom for context\u002FKV cache).\n2. Benchmark your hardware: try full GPU offload first; fall back to partial; tune thread count \n(start at physical core count).\n3. Set context size for your use case; enable KV cache quantization if context is the constraint.\n4. Run server mode for app integration; test concurrent requests and streaming.\n5. Tune batch parameters for your workload: prompt-heavy (bigger batches) vs. generation-heavy \n(smaller).\n6. Consider speculative decoding with a small draft model for interactive latency wins.\n\n```text\nTuning order (biggest wins first):\n1. GPU offload: max layers that fit VRAM\n2. Quant choice: biggest that fits RAM\n3. Threads: physical core count (CPU-bound parts)\n4. Speculative decoding: small draft model\n5. Batch\u002Fcontext knobs for your workload mix\n```\n\n## Common pitfalls\n\n- **Wrong thread count**: too many threads on hyperthreaded cores can slow things down. Start with \nphysical cores.\n- **Quant too aggressive**: Q2 fits but quality suffers. Q4_K_M is the default for a reason — \ndeviate deliberately.\n- **Ignoring KV cache**: long contexts eat RAM beyond the model weights. Budget for it or quantize \nthe cache.\n- **No GPU offload configured**: running full CPU when a GPU sits idle. Check offload actually \nengaged.\n- **Server defaults for production**: default concurrency and timeouts are demo settings. Tune for \nyour load.\n- **Model-architecture mismatch**: not every GGUF file works with every build. Match the engine \nversion to the model format.\n","# llama.cpp Guide\n\nllama.cpp is the portable inference engine: C++-based, runs everywhere from phones to servers, with \nthe best CPU inference in the business and broad GGUF model support. It's the power tool behind \nmost serious local LLM setups.\n\n## Overview\n\nThe mental model: models in GGUF format (quantized variants from 2-bit to 8-bit), an engine that \nsplits work between CPU and GPU layers, and two interfaces — a CLI for direct use and a server \nmode exposing an OpenAI-compatible API. Performance comes from tuning thread counts, batch sizes, \nand GPU offloading to your specific hardware. It's lower-level than packaged runners, which means \nmore control and more knobs.\n\n## When to use\n\n- Maximum local performance: squeezing the most tokens\u002Fsec from your hardware.\n- CPU-only inference where other engines struggle.\n- Cross-platform deployment: same engine on Mac, Linux, Windows, even mobile.\n- Fine-grained control: custom quantization, speculative decoding, grammar-constrained output.\n\n## Core concepts\n\n- **GGUF format**: the model container — architecture, weights, and metadata in one file, with \nquantization variants (Q4_K_M is the popular sweet spot). Choose the largest quant that fits your \nRAM.\n- **GPU offloading**: layers assigned to GPU (`-ngl`) vs. CPU. Full offload when VRAM fits; partial \noffload still beats CPU-only. More offloaded layers = faster.\n- **Threading and batching**: CPU thread count matched to your cores (physical, not hyperthreaded, \nusually); batch size tuned for prompt processing vs. generation.\n- **Server mode**: HTTP server with OpenAI-compatible endpoints — the integration point for apps \nand agents. Supports concurrent requests with continuous batching.\n- **Context management**: context size set at load (`-c`); KV cache quantization to stretch long \ncontexts into limited RAM.\n- **Advanced features**: speculative decoding (draft model proposes, main model verifies — big \nspeedups), grammar-constrained generation (JSON mode via grammars), multimodal support.\n\n## Practical workflow\n\n1. Pick the GGUF quant: largest that fits comfortably in RAM (leave headroom for context\u002FKV cache).\n2. Benchmark your hardware: try full GPU offload first; fall back to partial; tune thread count \n(start at physical core count).\n3. Set context size for your use case; enable KV cache quantization if context is the constraint.\n4. Run server mode for app integration; test concurrent requests and streaming.\n5. Tune batch parameters for your workload: prompt-heavy (bigger batches) vs. generation-heavy \n(smaller).\n6. Consider speculative decoding with a small draft model for interactive latency wins.\n\n```text\nTuning order (biggest wins first):\n1. GPU offload: max layers that fit VRAM\n2. Quant choice: biggest that fits RAM\n3. Threads: physical core count (CPU-bound parts)\n4. Speculative decoding: small draft model\n5. Batch\u002Fcontext knobs for your workload mix\n```\n\n## Common pitfalls\n\n- **Wrong thread count**: too many threads on hyperthreaded cores can slow things down. Start with \nphysical cores.\n- **Quant too aggressive**: Q2 fits but quality suffers. Q4_K_M is the default for a reason — \ndeviate deliberately.\n- **Ignoring KV cache**: long contexts eat RAM beyond the model weights. Budget for it or quantize \nthe cache.\n- **No GPU offload configured**: running full CPU when a GPU sits idle. Check offload actually \nengaged.\n- **Server defaults for production**: default concurrency and timeouts are demo settings. Tune for \nyour load.\n- **Model-architecture mismatch**: not every GGUF file works with every build. Match the engine \nversion to the model format.\n",{"name":5,"category":24,"description":7},"ai-research","main","af388a77c2fe5e0b4fd879cca4300a0ffa9a6579","skills\u002Fams-llamacpp-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 llamacpp-guide",{"tool":39,"command":40},"bunx","bunx skills add TanNhatCMS\u002Fjarvis --skill llamacpp-guide",{"tool":42,"command":43},"pnpm","pnpm dlx skills add TanNhatCMS\u002Fjarvis --skill llamacpp-guide","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fams-llamacpp-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",[],[]]