[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7EVAPPWX37B5EPV33NHX61F":3,"skill-same:01M7EVAPPWX37B5EPV33NHX61F":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},"01M7EVAPPWX37B5EPV33NHX61F","phi-guide",null,"Build with Microsoft's Phi small language models — compact reasoning models for edge and cost-efficient apps.","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-phi-guide\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: phi-guide\ndescription: Build with Microsoft's Phi small language models — compact reasoning models for edge and cost-efficient apps.\ncategory: ai-research\n---\n\n## Overview\n\nPhi (Microsoft) is a family of small language models (roughly 3B–14B) known for\npunching far above their weight — particularly in reasoning, math, and code —\nthanks to training on high-quality synthetic and curated data rather than raw\nscale. The \"textbook quality data\" story is the differentiator: Phi demonstrates\nthat data quality can substitute for parameters, producing small models with\nsurprising capability.\n\nFor builders, Phi's niche overlaps Gemma's (small, efficient, deployable) with\na reasoning emphasis: on-device copilots, cost-efficient RAG components,\nedge inference, and applications where a small model's reasoning ability\nmatters. Microsoft's backing means good tooling integration (ONNX, Windows,\nAzure) and continued development.\n\nThe practical stance: Phi belongs in every small-model evaluation, especially\nwhen the task involves reasoning rather than just fluency.\n\n## When to use\n\n- Small-model reasoning tasks: math, logic, code assistance at small scale.\n- On-device and edge AI (Windows, mobile, browsers).\n- Cost-efficient components in larger pipelines (classifiers, extractors,\n  rerankers' assistants).\n- Offline\u002Fair-gapped deployments needing capable small models.\n- Educational and research use of compact models.\n- Latency-critical inference where small models' speed matters.\n\n## Core concepts\n\n- **Data-quality thesis**: Phi's capability comes from curated\u002Fsynthetic\n  training data, not scale. The lesson generalizes: for fine-tuning small\n  models, invest in data quality over quantity.\n- **Size range**: roughly 3B–14B across generations. Benchmark the range —\n  newer generations improve the quality\u002Fsize tradeoff.\n- **Reasoning emphasis**: unusually strong reasoning for the size class. Test\n  on reasoning benchmarks relevant to your task, not just chat fluency.\n- **Microsoft ecosystem**: ONNX export, Windows AI integration, Azure\n  deployment paths. If you're Microsoft-centric, the integration is smooth.\n- **Instruction-tuned variants**: tuned for assistant behavior. Use IT\n  variants for interactive use; base for fine-tuning.\n- **Edge deployment**: small sizes + quantization = genuine on-device AI.\n  Validate on target hardware with your latency budgets.\n- **Multimodal variants**: vision-capable Phi versions for compact\n  image+text tasks. Evaluate against alternatives at the size.\n- **Licensing**: Microsoft's terms for Phi — read for your use case, as with\n  any model family.\n\n## Practical workflow\n\n1. **Benchmark reasoning at small scale.** Your reasoning tasks on Phi vs.\n   other small models (Gemma, Qwen small, Llama small). Phi's edge is\n   reasoning — verify it on your problems.\n2. **Test the size range.** Newer Phi generations shift the optimum. Find the\n   smallest adequate size for your task.\n3. **Evaluate deployment targets.** On-device: quantize and test on hardware.\n   Cloud: compare API\u002Fself-host economics at your volume.\n4. **Consider the Microsoft path.** If you're on Azure\u002FWindows: evaluate the\n   integrated deployment options — they may simplify operations.\n5. **Fine-tune where close.** Small models + quality data = effective\n   fine-tuning. If base Phi is near your bar, targeted training may close it.\n6. **Check licensing.** Confirm terms for your commercial or edge deployment\n   scenario.\n7. **Track generations.** Phi improves rapidly per generation; re-benchmark\n   on new releases.\n\nChecklist for Phi in production:\n- Reasoning advantage verified against small-model alternatives.\n- Smallest adequate size selected.\n- Edge deployment validated on target hardware (if applicable).\n- License confirmed for your scenario.\n- Fine-tuning evaluated where base is close.\n\n## Common pitfalls\n\n- **Fluency-only evals.** Testing chat smoothness and missing Phi's reasoning\n  differentiation. Evaluate reasoning specifically.\n- **Ignoring the data lesson.** Fine-tuning Phi on sloppy data and wondering\n  why it underperforms. The family's thesis is data quality — honor it.\n- **Generation blindness.** Running an old Phi when newer generations are\n  much better per parameter. Track releases.\n- **Edge assumptions.** Assuming on-device works without testing on the\n  actual hardware with real latency budgets.\n- **Wrong ecosystem.** Fighting the tooling when another small model has\n  better support for your stack. Match model to deployment reality.\n- **Overestimating small models.** Phi is impressive for its size, not\n  limitless. Honest task-difficulty assessment.\n- **Skipping quantization tests.** Edge deployment without validating\n  quantized quality on your evals.\n- **License not checked.** Assuming terms from the general \"open\" narrative.\n  Read the actual terms.\n","## Overview\n\nPhi (Microsoft) is a family of small language models (roughly 3B–14B) known for\npunching far above their weight — particularly in reasoning, math, and code —\nthanks to training on high-quality synthetic and curated data rather than raw\nscale. The \"textbook quality data\" story is the differentiator: Phi demonstrates\nthat data quality can substitute for parameters, producing small models with\nsurprising capability.\n\nFor builders, Phi's niche overlaps Gemma's (small, efficient, deployable) with\na reasoning emphasis: on-device copilots, cost-efficient RAG components,\nedge inference, and applications where a small model's reasoning ability\nmatters. Microsoft's backing means good tooling integration (ONNX, Windows,\nAzure) and continued development.\n\nThe practical stance: Phi belongs in every small-model evaluation, especially\nwhen the task involves reasoning rather than just fluency.\n\n## When to use\n\n- Small-model reasoning tasks: math, logic, code assistance at small scale.\n- On-device and edge AI (Windows, mobile, browsers).\n- Cost-efficient components in larger pipelines (classifiers, extractors,\n  rerankers' assistants).\n- Offline\u002Fair-gapped deployments needing capable small models.\n- Educational and research use of compact models.\n- Latency-critical inference where small models' speed matters.\n\n## Core concepts\n\n- **Data-quality thesis**: Phi's capability comes from curated\u002Fsynthetic\n  training data, not scale. The lesson generalizes: for fine-tuning small\n  models, invest in data quality over quantity.\n- **Size range**: roughly 3B–14B across generations. Benchmark the range —\n  newer generations improve the quality\u002Fsize tradeoff.\n- **Reasoning emphasis**: unusually strong reasoning for the size class. Test\n  on reasoning benchmarks relevant to your task, not just chat fluency.\n- **Microsoft ecosystem**: ONNX export, Windows AI integration, Azure\n  deployment paths. If you're Microsoft-centric, the integration is smooth.\n- **Instruction-tuned variants**: tuned for assistant behavior. Use IT\n  variants for interactive use; base for fine-tuning.\n- **Edge deployment**: small sizes + quantization = genuine on-device AI.\n  Validate on target hardware with your latency budgets.\n- **Multimodal variants**: vision-capable Phi versions for compact\n  image+text tasks. Evaluate against alternatives at the size.\n- **Licensing**: Microsoft's terms for Phi — read for your use case, as with\n  any model family.\n\n## Practical workflow\n\n1. **Benchmark reasoning at small scale.** Your reasoning tasks on Phi vs.\n   other small models (Gemma, Qwen small, Llama small). Phi's edge is\n   reasoning — verify it on your problems.\n2. **Test the size range.** Newer Phi generations shift the optimum. Find the\n   smallest adequate size for your task.\n3. **Evaluate deployment targets.** On-device: quantize and test on hardware.\n   Cloud: compare API\u002Fself-host economics at your volume.\n4. **Consider the Microsoft path.** If you're on Azure\u002FWindows: evaluate the\n   integrated deployment options — they may simplify operations.\n5. **Fine-tune where close.** Small models + quality data = effective\n   fine-tuning. If base Phi is near your bar, targeted training may close it.\n6. **Check licensing.** Confirm terms for your commercial or edge deployment\n   scenario.\n7. **Track generations.** Phi improves rapidly per generation; re-benchmark\n   on new releases.\n\nChecklist for Phi in production:\n- Reasoning advantage verified against small-model alternatives.\n- Smallest adequate size selected.\n- Edge deployment validated on target hardware (if applicable).\n- License confirmed for your scenario.\n- Fine-tuning evaluated where base is close.\n\n## Common pitfalls\n\n- **Fluency-only evals.** Testing chat smoothness and missing Phi's reasoning\n  differentiation. Evaluate reasoning specifically.\n- **Ignoring the data lesson.** Fine-tuning Phi on sloppy data and wondering\n  why it underperforms. The family's thesis is data quality — honor it.\n- **Generation blindness.** Running an old Phi when newer generations are\n  much better per parameter. Track releases.\n- **Edge assumptions.** Assuming on-device works without testing on the\n  actual hardware with real latency budgets.\n- **Wrong ecosystem.** Fighting the tooling when another small model has\n  better support for your stack. Match model to deployment reality.\n- **Overestimating small models.** Phi is impressive for its size, not\n  limitless. Honest task-difficulty assessment.\n- **Skipping quantization tests.** Edge deployment without validating\n  quantized quality on your evals.\n- **License not checked.** Assuming terms from the general \"open\" narrative.\n  Read the actual terms.\n",{"name":5,"category":24,"description":7},"ai-research","main","af388a77c2fe5e0b4fd879cca4300a0ffa9a6579","skills\u002Fams-phi-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 phi-guide",{"tool":39,"command":40},"bunx","bunx skills add TanNhatCMS\u002Fjarvis --skill phi-guide",{"tool":42,"command":43},"pnpm","pnpm dlx skills add TanNhatCMS\u002Fjarvis --skill phi-guide","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FTanNhatCMS\u002Fjarvis\u002Fblob\u002Fmain\u002Fskills\u002Fams-phi-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",[],[]]