Raymond D Field·2d agoAn encoder-decoder transformer from scratch, part 1 — PerlI have read many articles and attended many courses on how the basic building blocks of AI are actually built. All of them focus on the…
Ragunath S·Sep 14Tokens aren’t wordsA next-token engine is not a database with manners — token IDs, scale, and lookup vs generate.
Ragunath S·Sep 14Stop renting the kitchenA baby next-token engine in plain Python — no libraries, no API key.
InTechnology HitsbyViscousPotential·Aug 26More Data Made My Obsidian Model Worse6.6% character error rate. About 2x better than the prototype, on real handwriting, on a real tablet. That’s the model that came out of…
Mathew K Sojan·Aug 20Building a Resort AI Assistant with Salesforce Agentforce - Part 1: From Data Model to Room SearchWhat happens when an AI assistant doesn’t just answer questions, but actually interacts with Salesforce data? I decided to find out by…
Akshaya Jonnalagadda·Aug 4I Shrunk nanoGPT Below One Million ParametersWhat building my first transformer on a dying MacBook Air taught me about where parameters in a model actually go.
AI Engineering·May 24Build a Local AI Agent From Scratch: A Deep Dive Tutorial That Rejects Fast-Food Learning
Ji-hoon Na·Apr 19From Scratch vs Transfer Learning : Hitting the Ground Running from the MidpointSo, you’ve got a fancy new model and some data. Should you train it from a blank slate (Scratch) or borrow some “brainpower” from a…
Kaan·Apr 16Part 1 — Baseline GPT from scratchThis is Part 1 of the “Journey to becoming an LLM Expert” series. Full series index → [link]
Aranya Ray·Apr 3Micro-GPT2 — Building GPT-2 (124M) from Scratch in PyTorchEver wondered what actually happens inside GPT-2 — the model that kicked off the modern era of large language models?
