How I automated groundwater prediction using geo-resistivity data and scikit-learn without a single expert geophysicist in the loop Read more ›
A beautiful RSS reader with a clean three-pane interface, Google Reader-compatible API, full-text search, and dark mode. Read more ›
Why an M4 Max Mac Studio with 128GB of unified memory is the best local LLM workstation for developers - running Qwen3, Ling and Gemma with llama.cpp and MLX. Read more ›
An annual blogging event by Robert Birming where bloggers share others’ posts and help others discover cool individuals in the IndieWeb / small web community. Read more ›
The Markdown elements outlined in the original design document. Read more ›
If you told someone you were an “AI Engineer” a few years ago, they probably assumed you were elbow-deep in PyTorch, wrangling massive… Read more ›
GPT-2-style LLM built from scratch in C/CUDA with hand-written backprop, BPE tokenizer, FlashAttention, pretraining, and SFT. - JustVugg/nanoeuler Read more ›
And why scikit-learn’s LinearRegression() isn’t doing what you think it’s doing Read more ›
Tracy Durnell is shaking things up: I’ve done a purge of my feed reader (down to 80 feeds lol). Rather than defaulting to my phone, I’m recommitting to books as my after-lunch reading. She goes on to list some new rules to help decide what to read online. Readers of my blog may recall, I […] Read more ›
Announcing folk.zone, a collection of free, community-run internet services I'm building as an IndieWeb commons. Including Mastodon, WriteFreely, Forgejo, Pixelfed, Lemmy, and more. All self-hosted on hardware in my living space. This is infrastructure for the common folk, not for enterprise or scale, inspired by omg.lol and rooted in the IndieWeb and Fediverse principles. Read more ›
If you are going to go to the bother of fine tuning for trivial problems like subject classification then I think you'll find Scikit Learn with a SGDClassifier on 2-grams will do probably just as well and be under 1MB for the trained classifier. Read more ›
Understanding Neurons, Neural Networks, Neural Connections, Activation Functions & More Read more ›
Part 1 of a 4-part Kaggle series — Feature Engineering and Generalization in the Ames Housing Competition Read more ›
Notedog is a portable, private Markdown journal for Android with browser editing, optional Git backup, and no cloud lock-in. Read more ›
#Python Fix ARM test listing failures caused by torch import at collection time (#20284) Summary: The ~180 ARM test listings fail in fbcode//mode/opt during pytest collection with AttributeError: module 'torch' has no attribute '_utils'. Root cause: torch._utils is importable but is not bound as an attribute on the torch module, so importing torch._dynamo (pulled in via executorch.exir / torchao when a test module is collected) reads torch._utils directly and crashes the whole listing. This i... Read more ›
NEW RESOURCES Spotted in my RSS feeds: Bubbles. From the About page: “We monitor thousands of independent, personal blogs via RSS. Every new post appears on Bubbles automatically. Nobody submits individual links. […] Read more ›
Like many I‘ve been watching this latest trend to ban social media sites for under 16s. The first I saw of it was , but it seems . Keir Starmer wants to give kids their childhood back, which is a catchy slogan I suppose, but feels a little bit hollow to me. Yeah perhaps for some that live out in the countryside, you can imagine they will be able to get back to horse riding and hiking or whatever, but for most inner city kids, it‘s not like there are loads of things to do. I read an interview ... Read more ›
I did not choose DeepSeek because I think GPT-4 is bad. I chose it because I was building a free app, and free apps teach you what actually matters pretty fast. The question was simple: how do I keep sessions cheap enough that people can practice a lot without me lighting money on fire? The answer pushed me toward DeepSeek-V3 (and later R1 for specific tasks). The real constraint was volume The app is a conversation practice tool. People come in to rehearse hard talks, not to admire the model... Read more ›
Series — Fine-Tuning, Smallest to Largest: LoRA (1.5B) ← you are here In I fully fine-tuned a 270M model — updating every weight. That's fine for a tiny model. It gets painful as models grow, because full fine-tuning needs gradients and optimizer state for every parameter (~4× the model size in memory). So: what do you do when the model is too big to comfortably fine-tune all of? The idea behind LoRA LoRA (Low-Rank Adaptation) rests on one observation: the change fine-tuning makes to a weight... Read more ›
Short open reading frames (sORFs) within non-coding RNAs (ncRNAs) have arisen as a hidden layer of gene regulation, encoding small peptides that represent a new class of cancer regulators with diagnostic and therapeutic potential. However, inferring associations between sORFs to specific cancer types remains challenging and requires computational approaches for accurate prediction. Recently, the CoraL framework introduced the first computational approach for predicting cancer-associated pepti... Read more ›