Skip to content
YuhuanStudio

YuhuanStudio

AI & machine-learning research and development

Building open Traditional-Chinese NLP resources — models & datasets — and the infrastructure, agents and products of an AI-native studio. Since 2020.

Website Hugging Face Yunxin API YunUI Docs

English · 简体中文 · 繁體中文


YuhuanStudio does AI & ML research and development, with a focus on Traditional-Chinese (繁體中文) natural-language processing for Mandarin speakers everywhere. Around that research we build the working stack of an AI-native studio — a model gateway, on-device inference, agents, and a design system.

🚧 Some of it is open here; a lot is still in progress and not yet open-sourced. This page grows as we release more.

🔬 Research — Traditional-Chinese NLP · open

The open core of the studio — models and datasets on 🤗 Hugging Face → yuhuanstudio, plus the tooling on GitHub.

  • Yunmo — small language models pretrained from scratch for Traditional Chinese: Yunmo-V1 (0.2B), Yunmo_v0.1 / Yunmo_v0 (0.1B). 🤗 models
  • 15 Traditional-Chinese pretraining datasets — Wikipedia (wikipedia-pretrain-zh-tw, wikipedia-image-zh-tw), forum (PTT-pretrain-zhtw), web (c4_pretrain_zhtw), news (OpenNewsArchive_pretrain_zhtw), dictionary (twdict_pretrain), reasoning (gsm8k_zhtw), document classification (TWDOCI), and more. 🤗 datasets
  • WikiZH_Dataset — the Chinese-Wikipedia corpus toolkit behind those datasets: pretraining & multimodal set generation, Traditional↔Simplified conversion, quality filtering, one unified CLI. Python

🚀 Open-source projects

Project What it is
YunUI A quietly beautiful, versioned React 19 + Tailwind v4 design system — OKLCH design tokens, primitives, patterns and AI components in one package, already shipping in production. npm i @yuhuanowo/yunuiui.yuhuanstudio.com
Yunshu Fast, local, multimodal inference engine for Apple Silicon — OpenAI/Anthropic-compatible, with native streaming speech-to-speech via Qwen3-Omni. Text, vision, audio, image & embeddings on MLX. Python

☁️ Products & services · live, source not yet open

  • Yunxin — a cross-generational LLM API gateway: 24+ providers behind one API key, across Chat Completions / Responses / Messages formats. Live at api.yuhuanstudio.com.
  • More agents, apps and infra are in active development and not yet public — watch this space.

🗄️ Archive

  • AI-Friend (archived) — an early experiment: a local-LLM AI VTuber / companion with real-time TTS and VTube Studio avatar control. Kept as a development record.

🌏 Find us

Website Hugging Face GitHub YouTube Twitch Bilibili Discord Instagram Threads


Made with care by YuhuanStudio · Traditional-Chinese AI, researched & built · since 2020 · Contributions welcome 🌱

Pinned Loading

  1. WikiZH_Dataset WikiZH_Dataset Public

    維基百科中文語料處理工具,用於生成預訓練與多模態資料集,支援繁簡體轉換、高質量過濾、統一 CLI 入口 🚀

    Python 2

  2. YunUI YunUI Public

    A quietly beautiful, versioned React 19 + Tailwind v4 design system — OKLCH design tokens, primitives, patterns, and AI components in one package.

    TypeScript 1

  3. Yunshu Yunshu Public

    Fast local multimodal (omni) inference engine for Apple Silicon — OpenAI/Anthropic-compatible, with native streaming speech-to-speech via Qwen3-Omni. Text, vision, audio, image & embeddings on MLX.

    Python 1

  4. YunAudio YunAudio Public

    An audio router for macOS with its own virtual device, and a signal path that can be proved bit-exact.

    Swift 2

Repositories

Showing 7 of 7 repositories

Top languages

Loading…

Most used topics

Loading…