ML Engineer · Open-Source Contributor · Designer · Community Builder
HF Transformers contributor · 72 models on HF Hub · marimo-learn co-author · Twinkle AI brand designer
I'm an ML engineer and designer based in Taipei, Taiwan 🇹🇼, focused on making AI tools accessible to Traditional Chinese speakers through open-source contributions, translation, and community building.
I contribute to the Hugging Face ecosystem (Transformers, Optimum-Quanto, HF Course) and the marimo reactive notebook project. I'm also a core member and brand designer of Twinkle AI — Taiwan's largest Traditional Chinese LLM research community with 1,000+ Discord members.
| Project | Role | Links |
|---|---|---|
| Transformers | Contributor to the core ML library | Commits → |
| Optimum-Quanto | Contributed StableDiffusion quantization example | Repo → |
| HF Course | Official Traditional Chinese translator | Repo → |
| HF Hub | 72 models · 11 datasets · 18 Spaces | Profile → |
| Project | Role | Stars |
|---|---|---|
| marimo-learn | Co-author of official educational notebooks |  |
| marimo-LLMs-from-scratch | Bilingual (EN/繁中) interactive LLM tutorial |  |
| marimo core | Official Traditional Chinese translator · Credited in release notes (v0.9.31+) | Repo → |
Core member and complete brand designer for Taiwan's Traditional Chinese LLM research community.
| Role | Details |
|---|---|
| Brand & Design | Created the Formosan leopard cat mascot and full visual identity used across Discord, HF, GitHub, and events · Brand Assets → |
| Model Deployment | Converted gemma-3-4B-T1-it to GGUF format (90 downloads/month) · Deployed live demo on HF Spaces with ZeroGPU |
| Content & Education | Author on AILogora · Host of 熬夜書訪 (late-night study sessions) |
| Data Curation | Formosa-Vision and Traditional Chinese datasets for model training |
- 🧪 ZipQDoRA — DoRA + quantization for personalized image generation · Curated LoRA/DoRA finetuning datasets on HF Hub
- 🎤 Build with AI 2026 (GDG Taipei × Twinkle AI, Jan 2026) — Talk on solving notebook reproducibility with dataflow graphs (80+ attendees)
- 📖 《Inference Engineering》書評 — Deep-dive review and community discussion on LLM inference optimization (200+ views on AILogora)
- 🎨 GPT Image 1 & Gemini Image Edit — Built and deployed image generation/editing demos on HF Spaces
- 🌐 使用 LM Studio 執行 gemma-3-4B-T1-it — Beginner tutorial for local LLM deployment (190+ views on AILogora)
- 🎙️ Twinkle AI 熬夜書訪 — Regular host covering attention mechanisms, reactive notebooks, and Traditional Chinese ML topics
| Area | Skills |
|---|---|
| Efficient ML | LoRA · DoRA · Quantization (GGUF, Quanto) · Model Compression |
| LLMs & VLMs | Post-training · Serving (vLLM, SGLang, LM Studio) · Traditional Chinese NLP |
| Diffusion Models | SDXL fine-tuning · StableDiffusion · Image generation |
| Tools & Infra | PyTorch · Transformers · marimo · Cloudflare Workers |
| Design | Brand identity · Mascot design · Visual communication |
| Languages | Python · SQL · 繁體中文 · English |
I'm open to collaboration on Hugging Face ecosystem tools, reactive notebooks (marimo), Traditional Chinese LLMs, AI community branding, and bilingual open-source education.
