Can a Tiny LLM Beat A Supervised Model?
1. Introduction Can a compact, fine-tuned LLM outperform a lightweight supervised model at structured extraction—specifically, parsing recipes from noisy, real-world web pages? That’s the technical ch...
Experiments in Supervised Learning Part 4
Speeding up the Training Loop When developing and evaluating different machine learning models, rapid iteration is key—especially as experiments shift from feature engineering to focused model evaluat...
Experiments in Supervised Learning Part 3
In the previous post we built a simple model to extract recipe data from a web page. But when we ran it on an actual recipe, the results were pretty rough: it achieved just 65% accuracy, mislabeled so...
Experiments in Supervised Learning Part 2
First Results: Building and Running the Initial Model In the last post, we looked at how an old side project —Recipe Folder— left me with a rich but messy dataset of recipes labeled by real users. We ...
Experiments in Supervised Learning: Lessons from an old Recipe App
How to Do Supervised Learning: Lessons from an Old Recipe App *Heads Up : This is a technical post and requires enough understanding of programming to be relatively comfortable in Visual Studio Code T...