https://dnhkng.github.io/David Noel NgML, Biotech, Hardware, and Coordination Problems. Sometimes I write about hard problems and how to solve them. 2026-04-25T16:49:07+02:00 David Noel Ng https://dnhkng.github.io/ Jekyll © 2026 David Noel Ng /assets/img/favicons/favicon.ico /assets/img/favicons/favicon-96x96.png What 2x GH200 delivers: memory paths for LLM inference2026-04-25T00:00:00+02:00 2026-04-25T00:00:00+02:00 https://dnhkng.github.io/posts/gh200-benchmarking/ David Noel Ng Introduction This article is mostly for me, as a way to record the peculiarities of my server; but it might come in handy for the ~3 other people running a home Grace-Hopper server? In a previous post I tuned vLLM for MiniMax M2.1 on this 9,000 euro dual GH200 desktop. That work answered one practical question. It also left the larger one open: what does this machine deliver when an inference ... LLM Neuroanatomy III: Why RYS Works — The Language-Agnostic Middle2026-03-26T00:00:00+01:00 2026-04-20T13:43:50+02:00 https://dnhkng.github.io/posts/sapir-whorf/ David Noel Ng If you haven’t heard about the Sapir-Whorf hypothesis, don’t worry, I hadn’t either until I saw a comment on Twitter about my RYS part II article. Maciej Stachowiak dropped the comment regarding my small experiment comparing language vs topic in LLMs. The data suggested that the middle layers of Qwen3.5-27B were organised more by sentence topic than by language, with much less sensitivity to th... LLM Neuroanatomy II: Modern LLM Hacking and hints of a Universal Language?2026-03-22T00:00:00+01:00 2026-04-01T18:01:05+02:00 https://dnhkng.github.io/posts/rys-ii/ David Noel Ng In Part 1, I described how duplicating a block of seven middle layers in Qwen2-72B — no weight changes, no training — produced the #1 model on the HuggingFace Open LLM Leaderboard. The method, which I called RYS (Repeat Your Self), was discovered using nothing but hard math probes and EQ-Bench on a pair of RTX 4090s. That was mid-2024. Since then, a flood of strong open-source models has arriv... LLM Neuroanatomy: How I Topped the LLM Leaderboard Without Changing a Single Weight2026-03-10T00:00:00+01:00 2026-03-22T12:00:08+01:00 https://dnhkng.github.io/posts/rys/ David Noel Ng In mid-2024, the HuggingFace Open LLM Leaderboard was the Colosseum for Open-Weight AI. Thousands of models were battling it out, submitted by both well-funded labs with teams of PhDs and fine-tuning wizards creating fantastically named models (e.g. Nous-Hermes, Dolphin and NeuralBeagle14-7B…), fighting for the top spot across six benchmarks: IFEval, BBH, MATH Lvl 5, GPQA, MuSR, and MMLU-PRO.... Arrhenius Integrals, IR Lasers, and Cooking Proteins2026-02-25T10:00:00+01:00 2026-02-26T14:18:47+01:00 https://dnhkng.github.io/posts/arrhenius-integrals/ David Noel Ng Introduction May 2016, Munich. I had just joined NanoTemper Technologies as a Bioanalytics Scientist. If you aren’t familiar with NanoTemper, they build high-end biophysical instruments. At the time, their newest platform was the Prometheus — a machine that performed nanoDSF (Differential Scanning Fluorimetry) for helping develop new medicines. Developing antibody drugs is hard; even when you...