Writing

Two promised features shipped, one deferred, and ten small bugs that had to be fixed before adding anything new.

Walking through the full LoRA fine-tuning code — loading TinyLlama, configuring LoRA, training on the Alpaca dataset, and testing the result.

How LoRA works, how quantization reduces memory, and how QLoRA combines both for efficient fine-tuning.

How oil is priced through spot and futures markets, what moves oil prices, and what the spreads between benchmarks tell us about the market.

The fundamentals of fine-tuning a large language model — from data preparation and hyperparameters to the training loop and evaluation.

What is crude oil, what comes out of a barrel, who produces and consumes it, and how it moves around the world — the foundation for understanding oil prices.

A personal budgeting web app that reverses the typical spending-then-tracking loop. Plan your budget first, then track transactions against the plan with live variance.