Hugh Lin · 林耿塬

I build developer tools and agent workflows, and run reproducible language-model & quantization experiments on Apple Silicon.

Local LLM & MLX Lab

Project	Focus
tfm-quant-mlx	TimesFM 3.0 on MLX: weight quantization, long-context KV compression, and reproducible benchmarks.
qjl-mlx	QJL and TurboQuant KV-cache quantization with MLX and PyTorch reference checks.
minimind-autoresearch	Training and evaluation experiments for a 26M Chinese language model on Apple Silicon.
gpt2-from-scratch	GPT-2 Small (124M) training, LoRA, evaluation, and inference on an M5 MacBook Air.
gemma4-vs-qwen35-benchmark	Local LLM benchmark comparing Gemma 4 and Qwen 3.5 on 32GB M5 MacBook Air + Ollama.

Tools & Workflows

Project	What it does
my-skills	Agent skills for technical writing, Obsidian, Git, and Mermaid diagrams.
x-algorithm-simulator	Interactive simulator of X's public recommendation pipeline using local fixtures.
ascii-image-studio	Terminal-style ASCII art generator from images, written in Rust.
rex-docs-zh	Unofficial Chinese community translation of the Rex documentation.

Open-Source Contributions

• obsidianmd/pele — 7 merged PRs fixing diagram layout, label rendering, C4 detection, and resize redraws in Obsidian's Mermaid-to-SVG renderer.

Stack

Python · TypeScript · Rust · MLX · PyTorch · Ollama

---

_{For project questions, please open an issue in the relevant repository.}