Show HN: Logistic regression beats billion-parameter models at classification
Category: ai-ml
Tags: logistic-regression, text-classification, cpu-inference, pretrained-models, multilingual
Score: 6.5/10 (Innovation: 6, Technical: 6, Documentation: 8, Utility: 6)
Jeffy provides a collection of pretrained text classifiers using logistic regression on top of frozen embeddings, runnable on CPU, with 68 classifiers covering English and 51 other languages. It includes a training pipeline for custom classifiers and an HTTP serving interface, demonstrating that simple linear models can compete with billion-parameter models for classification tasks when paired with strong embeddings.
Target audience: backend devs, data scientists, ML engineers
Repository: https://jeffyclassify.com/benchmark/ · Python · MIT · 106 stars
View on Hacker News