AGENT SKILL

yi-guide

Build with 01.AI's Yi models — capable bilingual open models with strong Chinese-English performance.

License 许可
MIT
所属仓库
TanNhatCMS/jarvis
最近更新
2026-11-07
Stars
★ 0
Forks
1

01 前言 Front Matter

name
yi-guide
category
ai-research
description
Build with 01.AI's Yi models — capable bilingual open models with strong Chinese-English performance.

03 SKILL.md 正文

Overview

Yi (01.AI, founded by Kai-Fu Lee) is a bilingual open-model family with particularly strong Chinese-English performance — the Yi models debuted with impressive benchmark scores in their size classes and a focus on high-quality bilingual training. Available in multiple sizes including long-context variants, Yi serves builders needing strong Chinese language capability in an open model.

For builders, Yi's place is in the bilingual open-model evaluation set: if your application serves Chinese-speaking users, Yi deserves benchmarking alongside Qwen, DeepSeek, and other bilingual families. The models run on standard stacks and are hosted by inference providers.

The practical stance: evaluate Yi where bilingual quality matters. Include it in the comparison set; let your evals decide.

When to use

  • Bilingual Chinese-English applications needing strong Chinese quality.
  • Open-model evaluations for Chinese-language tasks.
  • Long-context bilingual document processing (Yi's long-context variants).
  • Fine-tuning bilingual base models.
  • Self-hosting capable bilingual open models.
  • Diversifying beyond the most common open families.

Core concepts

  • Bilingual training focus: strong Chinese and English from the training mix. For Chinese-language products, this is the relevant axis — test it directly.
  • Size options: multiple sizes across releases. Benchmark the ladder to right-size for your task.
  • Long-context variants: extended-context versions for document-scale bilingual work. Test long-context quality on your real documents.
  • Open weights: downloadable, self-hostable, fine-tunable. Check license terms per release.
  • Standard-stack compatibility: runs on vLLM, TGI, llama.cpp, and provider APIs. No special deployment requirements.
  • Instruction-tuned variants: chat-tuned versions for assistants and dialogue.
  • Vision extensions: multimodal Yi variants for image+text bilingual tasks — evaluate against alternatives on your images.
  • Release cadence: the family evolves; track new releases and re-benchmark deliberately.

Practical workflow

  1. Define bilingual requirements. Which tasks in Chinese, which in English, which mixed? The eval set should reflect the real language mix.
  2. Benchmark against bilingual peers. Yi vs. Qwen, DeepSeek, and others on your bilingual eval set. Per-language scores, not just aggregates.
  3. Test long-context if relevant. Your longest real bilingual documents; questions spanning the input.
  4. Right-size via the ladder. Smallest adequate size for your quality bar.
  5. Check licensing. Verify terms per release for your use case.
  6. Evaluate deployment. API for speed; self-hosted for volume economics.
  7. Pin versions; track releases. Production pins; deliberate re-evaluation on new releases.

Checklist for Yi in production:

  • Bilingual eval set reflecting your real language mix.
  • Benchmarked against bilingual peer models.
  • Long-context validated on real documents (if used).
  • License verified; version pinned.
  • Deployment economics modeled.

Common pitfalls

  • Aggregate-score blindness. A good average hiding weak Chinese (or English). Score per language.
  • Ignoring bilingual peers. Evaluating Yi alone instead of against Qwen, DeepSeek, and others. The decision is relative.
  • Long-context assumed. Extended window without testing quality over long bilingual inputs.
  • License not verified. Assuming terms across releases.
  • Wrong size. Not ladder-testing; over- or under-provisioning.
  • ** monolingual evals for bilingual products.** Testing only English for a Chinese-serving product (or vice versa).
  • Version drift. Floating references updating underneath production.
  • Static choice. Not re-evaluating as the bilingual open-model field advances.

04 同仓库更多 Skills

TanNhatCMS
TanNhatCMS/jarvis

web-scrape

Turn web pages into LLM-ready markdown via the Firecrawl API (renders JS, anti-bot handling). Use when browser_open/fetch fails on JS-heavy pages, or when you need clean markdown/structured extraction from a site. Requires FIRECRAWL_API_KEY.

★ 0
TanNhatCMS
TanNhatCMS/jarvis

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/CSS, 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.

★ 0
TanNhatCMS
TanNhatCMS/jarvis

test-driven-development

Use when implementing any feature or bugfix, before writing implementation code

★ 0
TanNhatCMS
TanNhatCMS/jarvis

systematic-debugging

Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes

★ 0
TanNhatCMS
TanNhatCMS/jarvis

requesting-code-review

Use when completing tasks, implementing major features, or before merging to verify work meets requirements

★ 0
TanNhatCMS
TanNhatCMS/jarvis

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).

★ 0