A from-scratch Go language LLM inference lab bench for playing with the mechanics of inference. It doesn’t use llama.cpp binding, as that’s a black box. The full inference loop written in Go so it’s open for exploration and modification. A fine-grain, thin-layer ggml binding provides GPU acceleration. Model architecture support is data-driven via toml definition files. It has a tool for generating SVG diagrams of the model architecture with layer details. While it doesn’t use llama.cpp for inference, there is tooling for validating the inference implemantation output against llama-server as a known working reference.

A guide for going from zero to personal site launch via Cloudflare and Github for just the cost of a domain. Includes source for a Cloudflare Worker project. Provides two options - minimal skeleton or complete Markdown -> HTML Web 1.0 workflow with sample content.

I can’t be the only one who wanted to do this.

a code generation tool that produces performant mobile, web, and desktop bindings from a single API definition.

C++ interaction engine and Vulkan renderer powering IKEA Kreativ for millions of mobile users

A Node.js-based glb model viewing and comparison utility derived from @shopify/screenshot-glb

Used internally at Geomagical/INGKA for validating compressed asset quality.

A performance experiment on Android using OpenGL ES 3.2 via Kotlin